chore(version): 1.4.8

- Added margin to the modal close button for improved spacing.
- Removed unnecessary divider in OCR settings for a cleaner layout.
- Set a minimum width for the search max result title to enhance alignment and readability.

.github/dependabot.yml

@@ -1,86 +1,17 @@
 version: 2
 updates:
-  - package-ecosystem: "npm"
-    directory: "/"
+  - package-ecosystem: 'github-actions'
+    directory: '/'
     schedule:
-      interval: "monthly"
-    open-pull-requests-limit: 7
-    target-branch: "main"
-    commit-message:
-      prefix: "chore"
-      include: "scope"
-    groups:
-      # 核心框架
-      core-framework:
-        patterns:
-          - "react"
-          - "react-dom"
-          - "electron"
-          - "typescript"
-          - "@types/react*"
-          - "@types/node"
-        update-types:
-          - "minor"
-          - "patch"
-
-      # Electron 生态和构建工具
-      electron-build:
-        patterns:
-          - "electron-*"
-          - "@electron*"
-          - "vite"
-          - "@vitejs/*"
-          - "dotenv-cli"
-          - "rollup-plugin-*"
-          - "@swc/*"
-        update-types:
-          - "minor"
-          - "patch"
-
-      # 测试工具
-      testing-tools:
-        patterns:
-          - "vitest"
-          - "@vitest/*"
-          - "playwright"
-          - "@playwright/*"
-          - "eslint*"
-          - "@eslint*"
-          - "prettier"
-          - "husky"
-          - "lint-staged"
-        update-types:
-          - "minor"
-          - "patch"
-
-      # CherryStudio 自定义包
-      cherrystudio-packages:
-        patterns:
-          - "@cherrystudio/*"
-        update-types:
-          - "minor"
-          - "patch"
-
-      # 兜底其他 dependencies
-      other-dependencies:
-        dependency-type: "production"
-
-      # 兜底其他 devDependencies
-      other-dev-dependencies:
-        dependency-type: "development"
-
-  - package-ecosystem: "github-actions"
-    directory: "/"
-    schedule:
-      interval: "weekly"
+      interval: 'monthly'
     open-pull-requests-limit: 3
     commit-message:
-      prefix: "ci"
-      include: "scope"
+      prefix: 'ci'
+      include: 'scope'
     groups:
       github-actions:
         patterns:
-          - "*"
+          - '*'
         update-types:
-          - "minor"
-          - "patch"
+          - 'minor'
+          - 'patch'

.github/workflows/pr-ci.yml

@@ -44,4 +44,4 @@ jobs:
         run: yarn build:check
 
       - name: Lint Check
-        run: yarn lint
+        run: yarn test:lint

.github/workflows/release.yml

@@ -27,7 +27,7 @@ jobs:
       - name: Check out Git repository
         uses: actions/checkout@v4
         with:
-          ref: main
+          fetch-depth: 0
 
       - name: Get release tag
         id: get-tag
@@ -79,6 +79,7 @@ jobs:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           RENDERER_VITE_AIHUBMIX_SECRET: ${{ vars.RENDERER_VITE_AIHUBMIX_SECRET }}
           NODE_OPTIONS: --max-old-space-size=8192
+          MAIN_VITE_MINERU_API_KEY: ${{ vars.MAIN_VITE_MINERU_API_KEY }}
 
       - name: Build Mac
         if: matrix.os == 'macos-latest'
@@ -95,6 +96,7 @@ jobs:
           RENDERER_VITE_AIHUBMIX_SECRET: ${{ vars.RENDERER_VITE_AIHUBMIX_SECRET }}
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           NODE_OPTIONS: --max-old-space-size=8192
+          MAIN_VITE_MINERU_API_KEY: ${{ vars.MAIN_VITE_MINERU_API_KEY }}
 
       - name: Build Windows
         if: matrix.os == 'windows-latest'
@@ -105,6 +107,7 @@ jobs:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           RENDERER_VITE_AIHUBMIX_SECRET: ${{ vars.RENDERER_VITE_AIHUBMIX_SECRET }}
           NODE_OPTIONS: --max-old-space-size=8192
+          MAIN_VITE_MINERU_API_KEY: ${{ vars.MAIN_VITE_MINERU_API_KEY }}
 
       - name: Release
         uses: ncipollo/release-action@v1
@@ -149,4 +152,4 @@ jobs:
           token: ${{ secrets.REPO_DISPATCH_TOKEN }}
           repository: CherryHQ/cherry-studio-docs
           event-type: update-download-version
-          client-payload: '{"version": "${{ steps.get-tag.outputs.tag }}"}'
+          client-payload: '{"version": "${{ steps.get-tag.outputs.tag }}"}'

.prettierrc

@@ -4,8 +4,5 @@
   "printWidth": 120,
   "trailingComma": "none",
   "endOfLine": "lf",
-  "bracketSameLine": true,
-  "tailwindStylesheet": "./src/renderer/src/assets/styles/tailwind.css",
-  "tailwindFunctions": ["clsx"],
-  "plugins": ["prettier-plugin-tailwindcss"],
+  "bracketSameLine": true
 }

.vscode/launch.json

@@ -7,7 +7,6 @@
       "request": "launch",
       "cwd": "${workspaceRoot}",
       "runtimeExecutable": "${workspaceRoot}/node_modules/.bin/electron-vite",
-      "runtimeVersion": "20",
       "windows": {
         "runtimeExecutable": "${workspaceRoot}/node_modules/.bin/electron-vite.cmd"
       },

.vscode/settings.json

@@ -1,7 +1,8 @@
 {
   "editor.formatOnSave": true,
   "editor.codeActionsOnSave": {
-    "source.fixAll.eslint": "explicit"
+    "source.fixAll.eslint": "explicit",
+    "source.organizeImports": "never"
   },
   "search.exclude": {
     "**/dist/**": true,

.yarn/patches/antd-npm-5.24.7-356a553ae5.patch

@@ -0,0 +1,69 @@
+diff --git a/es/dropdown/dropdown.js b/es/dropdown/dropdown.js
+index 986877a762b9ad0aca596a8552732cd12d2eaabb..1f18aa2ea745e68950e4cee16d4d655f5c835fd5 100644
+--- a/es/dropdown/dropdown.js
++++ b/es/dropdown/dropdown.js
+@@ -2,7 +2,7 @@
+ 
+ import * as React from 'react';
+ import LeftOutlined from "@ant-design/icons/es/icons/LeftOutlined";
+-import RightOutlined from "@ant-design/icons/es/icons/RightOutlined";
++import { ChevronRight } from 'lucide-react';
+ import classNames from 'classnames';
+ import RcDropdown from 'rc-dropdown';
+ import useEvent from "rc-util/es/hooks/useEvent";
+@@ -158,8 +158,10 @@ const Dropdown = props => {
+         className: `${prefixCls}-menu-submenu-arrow`
+       }, direction === 'rtl' ? (/*#__PURE__*/React.createElement(LeftOutlined, {
+         className: `${prefixCls}-menu-submenu-arrow-icon`
+-      })) : (/*#__PURE__*/React.createElement(RightOutlined, {
+-        className: `${prefixCls}-menu-submenu-arrow-icon`
++      })) : (/*#__PURE__*/React.createElement(ChevronRight, {
++        size: 16,
++        strokeWidth: 1.8,
++        className: `${prefixCls}-menu-submenu-arrow-icon lucide-custom`
+       }))),
+       mode: "vertical",
+       selectable: false,
+diff --git a/es/dropdown/style/index.js b/es/dropdown/style/index.js
+index 768c01783002c6901c85a73061ff6b3e776a60ce..39b1b95a56cdc9fb586a193c3adad5141f5cf213 100644
+--- a/es/dropdown/style/index.js
++++ b/es/dropdown/style/index.js
+@@ -240,7 +240,8 @@ const genBaseStyle = token => {
+               marginInlineEnd: '0 !important',
+               color: token.colorTextDescription,
+               fontSize: fontSizeIcon,
+-              fontStyle: 'normal'
++              fontStyle: 'normal',
++              marginTop: 3,
+             }
+           }
+         }),
+diff --git a/es/select/useIcons.js b/es/select/useIcons.js
+index 959115be936ef8901548af2658c5dcfdc5852723..c812edd52123eb0faf4638b1154fcfa1b05b513b 100644
+--- a/es/select/useIcons.js
++++ b/es/select/useIcons.js
+@@ -4,10 +4,10 @@ import * as React from 'react';
+ import CheckOutlined from "@ant-design/icons/es/icons/CheckOutlined";
+ import CloseCircleFilled from "@ant-design/icons/es/icons/CloseCircleFilled";
+ import CloseOutlined from "@ant-design/icons/es/icons/CloseOutlined";
+-import DownOutlined from "@ant-design/icons/es/icons/DownOutlined";
+ import LoadingOutlined from "@ant-design/icons/es/icons/LoadingOutlined";
+ import SearchOutlined from "@ant-design/icons/es/icons/SearchOutlined";
+ import { devUseWarning } from '../_util/warning';
++import { ChevronDown } from 'lucide-react';
+ export default function useIcons(_ref) {
+   let {
+     suffixIcon,
+@@ -56,8 +56,10 @@ export default function useIcons(_ref) {
+           className: iconCls
+         }));
+       }
+-      return getSuffixIconNode(/*#__PURE__*/React.createElement(DownOutlined, {
+-        className: iconCls
++      return getSuffixIconNode(/*#__PURE__*/React.createElement(ChevronDown, {
++        size: 16,
++        strokeWidth: 1.8,
++        className: `${iconCls} lucide-custom`
+       }));
+     };
+   }

.yarn/patches/app-builder-lib-npm-26.0.15-360e5b0476.patch

@@ -65,11 +65,44 @@ index e8bd7bb46c8a54b3f55cf3a853ef924195271e01..f956e9f3fe9eb903c78aef3502553b01
              await packager.info.emitArtifactBuildCompleted({
                  file: installerPath,
                  updateInfo,
+diff --git a/out/util/yarn.js b/out/util/yarn.js
+index 1ee20f8b252a8f28d0c7b103789cf0a9a427aec1..c2878ec54d57da50bf14225e0c70c9c88664eb8a 100644
+--- a/out/util/yarn.js
++++ b/out/util/yarn.js
+@@ -140,6 +140,7 @@ async function rebuild(config, { appDir, projectDir }, options) {
+         arch,
+         platform,
+         buildFromSource,
++        ignoreModules: config.excludeReBuildModules || undefined,
+         projectRootPath: projectDir,
+         mode: config.nativeRebuilder || "sequential",
+         disablePreGypCopy: true,
 diff --git a/scheme.json b/scheme.json
-index 433e2efc9cef156ff5444f0c4520362ed2ef9ea7..a89c7a9b0b608fef67902c49106a43ebd0fa8b61 100644
+index 433e2efc9cef156ff5444f0c4520362ed2ef9ea7..0167441bf928a92f59b5dbe70b2317a74dda74c9 100644
 --- a/scheme.json
 +++ b/scheme.json
-@@ -1975,6 +1975,13 @@
+@@ -1825,6 +1825,20 @@
+             "string"
+           ]
+         },
++        "excludeReBuildModules": {
++          "anyOf": [
++            {
++              "items": {
++                "type": "string"
++              },
++              "type": "array"
++            },
++            {
++              "type": "null"
++            }
++          ],
++          "description": "The modules to exclude from the rebuild."
++        },
+         "executableArgs": {
+           "anyOf": [
+             {
+@@ -1975,6 +1989,13 @@
            ],
            "description": "The mime types in addition to specified in the file associations. Use it if you don't want to register a new mime type, but reuse existing."
          },
@@ -83,7 +116,7 @@ index 433e2efc9cef156ff5444f0c4520362ed2ef9ea7..a89c7a9b0b608fef67902c49106a43eb
          "packageCategory": {
            "description": "backward compatibility + to allow specify fpm-only category for all possible fpm targets in one place",
            "type": [
-@@ -2327,6 +2334,13 @@
+@@ -2327,6 +2348,13 @@
      "MacConfiguration": {
        "additionalProperties": false,
        "properties": {
@@ -97,7 +130,28 @@ index 433e2efc9cef156ff5444f0c4520362ed2ef9ea7..a89c7a9b0b608fef67902c49106a43eb
          "additionalArguments": {
            "anyOf": [
              {
-@@ -2737,7 +2751,7 @@
+@@ -2527,6 +2555,20 @@
+             "string"
+           ]
+         },
++        "excludeReBuildModules": {
++          "anyOf": [
++            {
++              "items": {
++                "type": "string"
++              },
++              "type": "array"
++            },
++            {
++              "type": "null"
++            }
++          ],
++          "description": "The modules to exclude from the rebuild."
++        },
+         "executableName": {
+           "description": "The executable name. Defaults to `productName`.",
+           "type": [
+@@ -2737,7 +2779,7 @@
            "type": "boolean"
          },
          "minimumSystemVersion": {
@@ -106,7 +160,7 @@ index 433e2efc9cef156ff5444f0c4520362ed2ef9ea7..a89c7a9b0b608fef67902c49106a43eb
            "type": [
              "null",
              "string"
-@@ -2959,6 +2973,13 @@
+@@ -2959,6 +3001,13 @@
      "MasConfiguration": {
        "additionalProperties": false,
        "properties": {
@@ -120,7 +174,28 @@ index 433e2efc9cef156ff5444f0c4520362ed2ef9ea7..a89c7a9b0b608fef67902c49106a43eb
          "additionalArguments": {
            "anyOf": [
              {
-@@ -3369,7 +3390,7 @@
+@@ -3159,6 +3208,20 @@
+             "string"
+           ]
+         },
++        "excludeReBuildModules": {
++          "anyOf": [
++            {
++              "items": {
++                "type": "string"
++              },
++              "type": "array"
++            },
++            {
++              "type": "null"
++            }
++          ],
++          "description": "The modules to exclude from the rebuild."
++        },
+         "executableName": {
+           "description": "The executable name. Defaults to `productName`.",
+           "type": [
+@@ -3369,7 +3432,7 @@
            "type": "boolean"
          },
          "minimumSystemVersion": {
@@ -129,7 +204,28 @@ index 433e2efc9cef156ff5444f0c4520362ed2ef9ea7..a89c7a9b0b608fef67902c49106a43eb
            "type": [
              "null",
              "string"
-@@ -6507,6 +6528,13 @@
+@@ -6381,6 +6444,20 @@
+             "string"
+           ]
+         },
++        "excludeReBuildModules": {
++          "anyOf": [
++            {
++              "items": {
++                "type": "string"
++              },
++              "type": "array"
++            },
++            {
++              "type": "null"
++            }
++          ],
++          "description": "The modules to exclude from the rebuild."
++        },
+         "executableName": {
+           "description": "The executable name. Defaults to `productName`.",
+           "type": [
+@@ -6507,6 +6584,13 @@
              "string"
            ]
          },
@@ -143,7 +239,28 @@ index 433e2efc9cef156ff5444f0c4520362ed2ef9ea7..a89c7a9b0b608fef67902c49106a43eb
          "protocols": {
            "anyOf": [
              {
-@@ -7376,6 +7404,13 @@
+@@ -7153,6 +7237,20 @@
+         "string"
+       ]
+     },
++    "excludeReBuildModules": {
++      "anyOf": [
++        {
++          "items": {
++            "type": "string"
++          },
++          "type": "array"
++        },
++        {
++          "type": "null"
++        }
++      ],
++      "description": "The modules to exclude from the rebuild."
++    },
+     "executableName": {
+       "description": "The executable name. Defaults to `productName`.",
+       "type": [
+@@ -7376,6 +7474,13 @@
        ],
        "description": "MAS (Mac Application Store) development options (`mas-dev` target)."
      },

.yarn/patches/openai-npm-5.1.0-0e7b3ccb07.patch

@@ -151,7 +151,7 @@ index 2404264d4ba0204322548945ebb7eab3bea82173..8f1bc45cc45e0797d50989d96b51147b
 +			"embeddings/decoding base64 embeddings from base64"
 +		);
 +		return response._thenUnwrap((response) => {
-+			if (response && response.data) {
++			if (response && response.data && typeof response.data[0]?.embedding === 'string') {
 +				response.data.forEach((embeddingBase64Obj) => {
 +					const embeddingBase64Str = embeddingBase64Obj.embedding;
 +					embeddingBase64Obj.embedding = (0, utils_1.toFloat32Array)(
@@ -266,7 +266,7 @@ index 19dcaef578c194a89759c4360073cfd4f7dd2cbf..0284e9cc615c900eff508eb595f7360a
 +			"embeddings/decoding base64 embeddings from base64"
 +		);
 +		return response._thenUnwrap((response) => {
-+			if (response && response.data) {
++			if (response && response.data && typeof response.data[0]?.embedding === 'string') {
 +				response.data.forEach((embeddingBase64Obj) => {
 +					const embeddingBase64Str = embeddingBase64Obj.embedding;
 +					embeddingBase64Obj.embedding = toFloat32Array(embeddingBase64Str);

README.md

@@ -1,3 +1,33 @@
+<div align="right" >
+  <details>
+    <summary >🌐 Language</summary>
+    <div>
+      <div align="right">
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=en">English</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=zh-CN">简体中文</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=zh-TW">繁體中文</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=ja">日本語</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=ko">한국어</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=hi">हिन्दी</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=th">ไทย</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=fr">Français</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=de">Deutsch</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=es">Español</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=it">Itapano</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=ru">Русский</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=pt">Português</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=nl">Nederlands</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=pl">Polski</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=ar">العربية</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=fa">فارسی</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=tr">Türkçe</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=vi">Tiếng Việt</a></p>
+        <p><a href="https://openaitx.github.io/view.html?user=CherryHQ&project=cherry-studio&lang=id">Bahasa Indonesia</a></p>
+      </div>
+    </div>
+  </details>
+</div>
+
 <h1 align="center">
   <a href="https://github.com/CherryHQ/cherry-studio/releases">
     <img src="https://github.com/CherryHQ/cherry-studio/blob/main/build/icon.png?raw=true" width="150" height="150" alt="banner" /><br>
@@ -167,6 +197,78 @@ For more detailed guidelines, please refer to our [Contributing Guide](./CONTRIB
 
 Thank you for your support and contributions!
 
+# 🔧 Developer Co-creation Program
+
+We are launching the Cherry Studio Developer Co-creation Program to foster a healthy and positive-feedback loop within the open-source ecosystem. We believe that great software is built collaboratively, and every merged pull request breathes new life into the project.
+
+We sincerely invite you to join our ranks of contributors and shape the future of Cherry Studio with us.
+
+## Contributor Rewards Program
+
+To give back to our core contributors and create a virtuous cycle, we have established the following long-term incentive plan.
+
+**The inaugural tracking period for this program will be Q3 2025 (July, August, September). Rewards for this cycle will be distributed on October 1st.**
+
+Within any tracking period (e.g., July 1st to September 30th for the first cycle), any developer who contributes more than **30 meaningful commits** to any of Cherry Studio's open-source projects on GitHub is eligible for the following benefits:
+
+- **Cursor Subscription Sponsorship**: Receive a **$70 USD** credit or reimbursement for your [Cursor](https://cursor.sh/) subscription, making AI your most efficient coding partner.
+- **Unlimited Model Access**: Get **unlimited** API calls for the **DeepSeek** and **Qwen** models.
+- **Cutting-Edge Tech Access**: Enjoy occasional perks, including API access to models like **Claude**, **Gemini**, and **OpenAI**, keeping you at the forefront of technology.
+
+## Growing Together & Future Plans
+
+A vibrant community is the driving force behind any sustainable open-source project. As Cherry Studio grows, so will our rewards program. We are committed to continuously aligning our benefits with the best-in-class tools and resources in the industry. This ensures our core contributors receive meaningful support, creating a positive cycle where developers, the community, and the project grow together.
+
+**Moving forward, the project will also embrace an increasingly open stance to give back to the entire open-source community.**
+
+## How to Get Started?
+
+We look forward to your first Pull Request!
+
+You can start by exploring our repositories, picking up a `good first issue`, or proposing your own enhancements. Every commit is a testament to the spirit of open source.
+
+Thank you for your interest and contributions.
+
+Let's build together.
+
+# 🏢 Enterprise Edition
+
+Building on the Community Edition, we are proud to introduce **Cherry Studio Enterprise Edition**—a privately deployable AI productivity and management platform designed for modern teams and enterprises.
+
+The Enterprise Edition addresses core challenges in team collaboration by centralizing the management of AI resources, knowledge, and data. It empowers organizations to enhance efficiency, foster innovation, and ensure compliance, all while maintaining 100% control over their data in a secure environment.
+
+## Core Advantages
+
+- **Unified Model Management**: Centrally integrate and manage various cloud-based LLMs (e.g., OpenAI, Anthropic, Google Gemini) and locally deployed private models. Employees can use them out-of-the-box without individual configuration.
+- **Enterprise-Grade Knowledge Base**: Build, manage, and share team-wide knowledge bases. Ensure knowledge is retained and consistent, enabling team members to interact with AI based on unified and accurate information.
+- **Fine-Grained Access Control**: Easily manage employee accounts and assign role-based permissions for different models, knowledge bases, and features through a unified admin backend.
+- **Fully Private Deployment**: Deploy the entire backend service on your on-premises servers or private cloud, ensuring your data remains 100% private and under your control to meet the strictest security and compliance standards.
+- **Reliable Backend Services**: Provides stable API services, enterprise-grade data backup and recovery mechanisms to ensure business continuity.
+
+## ✨ Online Demo
+
+> 🚧 **Public Beta Notice**
+>
+> The Enterprise Edition is currently in its early public beta stage, and we are actively iterating and optimizing its features. We are aware that it may not be perfectly stable yet. If you encounter any issues or have valuable suggestions during your trial, we would be very grateful if you could contact us via email to provide feedback.
+
+**🔗 [Cherry Studio Enterprise](https://www.cherry-ai.com/enterprise)**
+
+## Version Comparison
+
+| Feature           | Community Edition                          | Enterprise Edition                                                                                                                      |
+| :---------------- | :----------------------------------------- | :-------------------------------------------------------------------------------------------------------------------------------------- |
+| **Open Source**   | ✅ Yes                                     | ⭕️ part. released to cust.                                                                                                             |
+| **Cost**          | Free for Personal Use / Commercial License | Buyout / Subscription Fee                                                                                                               |
+| **Admin Backend** | —                                          | ● Centralized **Model** Access<br>● **Employee** Management<br>● Shared **Knowledge Base**<br>● **Access** Control<br>● **Data** Backup |
+| **Server**        | —                                          | ✅ Dedicated Private Deployment                                                                                                         |
+
+## Get the Enterprise Edition
+
+We believe the Enterprise Edition will become your team's AI productivity engine. If you are interested in Cherry Studio Enterprise Edition and would like to learn more, request a quote, or schedule a demo, please contact us.
+
+- **For Business Inquiries & Purchasing**:
+  **📧 [bd@cherry-ai.com](mailto:bd@cherry-ai.com)**
+
 # 🔗 Related Projects
 
 - [one-api](https://github.com/songquanpeng/one-api):LLM API management and distribution system, supporting mainstream models like OpenAI, Azure, and Anthropic. Features unified API interface, suitable for key management and secondary distribution.
@@ -185,16 +287,18 @@ Thank you for your support and contributions!
 [![Star History Chart](https://api.star-history.com/svg?repos=CherryHQ/cherry-studio&type=Timeline)](https://star-history.com/#CherryHQ/cherry-studio&Timeline)
 
 <!-- Links & Images -->
+
 [deepwiki-shield]: https://img.shields.io/badge/Deepwiki-CherryHQ-0088CC?style=plastic
 [deepwiki-link]: https://deepwiki.com/CherryHQ/cherry-studio
 [twitter-shield]: https://img.shields.io/badge/Twitter-CherryStudioApp-0088CC?style=plastic&logo=x
-[twitter-link]: https://twitter.com/CherryStudioApp
+[twitter-link]: https://twitter.com/CherryStudioHQ
 [discord-shield]: https://img.shields.io/badge/Discord-@CherryStudio-0088CC?style=plastic&logo=discord
 [discord-link]: https://discord.gg/wez8HtpxqQ
 [telegram-shield]: https://img.shields.io/badge/Telegram-@CherryStudioAI-0088CC?style=plastic&logo=telegram
 [telegram-link]: https://t.me/CherryStudioAI
 
 <!-- Links & Images -->
+
 [github-stars-shield]: https://img.shields.io/github/stars/CherryHQ/cherry-studio?style=social
 [github-stars-link]: https://github.com/CherryHQ/cherry-studio/stargazers
 [github-forks-shield]: https://img.shields.io/github/forks/CherryHQ/cherry-studio?style=social
@@ -205,6 +309,7 @@ Thank you for your support and contributions!
 [github-contributors-link]: https://github.com/CherryHQ/cherry-studio/graphs/contributors
 
 <!-- Links & Images -->
+
 [license-shield]: https://img.shields.io/badge/License-AGPLv3-important.svg?style=plastic&logo=gnu
 [license-link]: https://www.gnu.org/licenses/agpl-3.0
 [commercial-shield]: https://img.shields.io/badge/License-Contact-white.svg?style=plastic&logoColor=white&logo=telegram&color=blue

components.json

@@ -1,21 +0,0 @@
-{
-  "$schema": "https://ui.shadcn.com/schema.json",
-  "style": "new-york",
-  "rsc": false,
-  "tsx": true,
-  "tailwind": {
-    "config": "",
-    "css": "src/renderer/src/assets/styles/tailwind.css",
-    "baseColor": "zinc",
-    "cssVariables": true,
-    "prefix": ""
-  },
-  "aliases": {
-    "components": "@renderer/ui/third-party",
-    "utils": "@renderer/utils",
-    "ui": "@renderer/ui",
-    "lib": "@renderer/lib",
-    "hooks": "@renderer/hooks"
-  },
-  "iconLibrary": "lucide"
-}

docs/README.ja.md

@@ -190,7 +190,7 @@ Cherry Studio への貢献を歓迎します！以下の方法で貢献できま
 [deepwiki-shield]: https://img.shields.io/badge/Deepwiki-CherryHQ-0088CC?style=plastic
 [deepwiki-link]: https://deepwiki.com/CherryHQ/cherry-studio
 [twitter-shield]: https://img.shields.io/badge/Twitter-CherryStudioApp-0088CC?style=plastic&logo=x
-[twitter-link]: https://twitter.com/CherryStudioApp
+[twitter-link]: https://twitter.com/CherryStudioHQ
 [discord-shield]: https://img.shields.io/badge/Discord-@CherryStudio-0088CC?style=plastic&logo=discord
 [discord-link]: https://discord.gg/wez8HtpxqQ
 [telegram-shield]: https://img.shields.io/badge/Telegram-@CherryStudioAI-0088CC?style=plastic&logo=telegram

docs/README.zh.md

@@ -202,7 +202,7 @@ https://docs.cherry-ai.com
 [deepwiki-shield]: https://img.shields.io/badge/Deepwiki-CherryHQ-0088CC?style=plastic
 [deepwiki-link]: https://deepwiki.com/CherryHQ/cherry-studio
 [twitter-shield]: https://img.shields.io/badge/Twitter-CherryStudioApp-0088CC?style=plastic&logo=x
-[twitter-link]: https://twitter.com/CherryStudioApp
+[twitter-link]: https://twitter.com/CherryStudioHQ
 [discord-shield]: https://img.shields.io/badge/Discord-@CherryStudio-0088CC?style=plastic&logo=discord
 [discord-link]: https://discord.gg/wez8HtpxqQ
 [telegram-shield]: https://img.shields.io/badge/Telegram-@CherryStudioAI-0088CC?style=plastic&logo=telegram

docs/technical/how-to-write-middlewares.md

@@ -0,0 +1,214 @@
+# 如何为 AI Provider 编写中间件
+
+本文档旨在指导开发者如何为我们的 AI Provider 框架创建和集成自定义中间件。中间件提供了一种强大而灵活的方式来增强、修改或观察 Provider 方法的调用过程，例如日志记录、缓存、请求/响应转换、错误处理等。
+
+## 架构概览
+
+我们的中间件架构借鉴了 Redux 的三段式设计，并结合了 JavaScript Proxy 来动态地将中间件应用于 Provider 的方法。
+
+- **Proxy**: 拦截对 Provider 方法的调用，并将调用引导至中间件链。
+- **中间件链**: 一系列按顺序执行的中间件函数。每个中间件都可以处理请求/响应，然后将控制权传递给链中的下一个中间件，或者在某些情况下提前终止链。
+- **上下文 (Context)**: 一个在中间件之间传递的对象，携带了关于当前调用的信息（如方法名、原始参数、Provider 实例、以及中间件自定义的数据）。
+
+## 中间件的类型
+
+目前主要支持两种类型的中间件，它们共享相似的结构但针对不同的场景：
+
+1.  **`CompletionsMiddleware`**: 专门为 `completions` 方法设计。这是最常用的中间件类型，因为它允许对 AI 模型的核心聊天/文本生成功能进行精细控制。
+2.  **`ProviderMethodMiddleware`**: 通用中间件，可以应用于 Provider 上的任何其他方法（例如，`translate`, `summarize` 等，如果这些方法也通过中间件系统包装）。
+
+## 编写一个 `CompletionsMiddleware`
+
+`CompletionsMiddleware` 的基本签名（TypeScript 类型）如下：
+
+```typescript
+import { AiProviderMiddlewareCompletionsContext, CompletionsParams, MiddlewareAPI } from './AiProviderMiddlewareTypes' // 假设类型定义文件路径
+
+export type CompletionsMiddleware = (
+  api: MiddlewareAPI<AiProviderMiddlewareCompletionsContext, [CompletionsParams]>
+) => (
+  next: (context: AiProviderMiddlewareCompletionsContext, params: CompletionsParams) => Promise<any> // next 返回 Promise<any> 代表原始SDK响应或下游中间件的结果
+) => (context: AiProviderMiddlewareCompletionsContext, params: CompletionsParams) => Promise<void> // 最内层函数通常返回 Promise<void>，因为结果通过 onChunk 或 context 副作用传递
+```
+
+让我们分解这个三段式结构：
+
+1.  **第一层函数 `(api) => { ... }`**:
+
+    - 接收一个 `api` 对象。
+    - `api` 对象提供了以下方法：
+      - `api.getContext()`: 获取当前调用的上下文对象 (`AiProviderMiddlewareCompletionsContext`)。
+      - `api.getOriginalArgs()`: 获取传递给 `completions` 方法的原始参数数组 (即 `[CompletionsParams]`)。
+      - `api.getProviderId()`: 获取当前 Provider 的 ID。
+      - `api.getProviderInstance()`: 获取原始的 Provider 实例。
+    - 此函数通常用于进行一次性的设置或获取所需的服务/配置。它返回第二层函数。
+
+2.  **第二层函数 `(next) => { ... }`**:
+
+    - 接收一个 `next` 函数。
+    - `next` 函数代表了中间件链中的下一个环节。调用 `next(context, params)` 会将控制权传递给下一个中间件，或者如果当前中间件是链中的最后一个，则会调用核心的 Provider 方法逻辑 (例如，实际的 SDK 调用)。
+    - `next` 函数接收当前的 `context` 和 `params` (这些可能已被上游中间件修改)。
+    - **重要的是**：`next` 的返回类型通常是 `Promise<any>`。对于 `completions` 方法，如果 `next` 调用了实际的 SDK，它将返回原始的 SDK 响应（例如，OpenAI 的流对象或 JSON 对象）。你需要处理这个响应。
+    - 此函数返回第三层（也是最核心的）函数。
+
+3.  **第三层函数 `(context, params) => { ... }`**:
+    - 这是执行中间件主要逻辑的地方。
+    - 它接收当前的 `context` (`AiProviderMiddlewareCompletionsContext`) 和 `params` (`CompletionsParams`)。
+    - 在此函数中，你可以：
+      - **在调用 `next` 之前**:
+        - 读取或修改 `params`。例如，添加默认参数、转换消息格式。
+        - 读取或修改 `context`。例如，设置一个时间戳用于后续计算延迟。
+        - 执行某些检查，如果不满足条件，可以不调用 `next` 而直接返回或抛出错误（例如，参数校验失败）。
+      - **调用 `await next(context, params)`**:
+        - 这是将控制权传递给下游的关键步骤。
+        - `next` 的返回值是原始的 SDK 响应或下游中间件的结果，你需要根据情况处理它（例如，如果是流，则开始消费流）。
+      - **在调用 `next` 之后**:
+        - 处理 `next` 的返回结果。例如，如果 `next` 返回了一个流，你可以在这里开始迭代处理这个流，并通过 `context.onChunk` 发送数据块。
+        - 基于 `context` 的变化或 `next` 的结果执行进一步操作。例如，计算总耗时、记录日志。
+        - 修改最终结果（尽管对于 `completions`，结果通常通过 `onChunk` 副作用发出）。
+
+### 示例：一个简单的日志中间件
+
+```typescript
+import {
+  AiProviderMiddlewareCompletionsContext,
+  CompletionsParams,
+  MiddlewareAPI,
+  OnChunkFunction // 假设 OnChunkFunction 类型被导出
+} from './AiProviderMiddlewareTypes' // 调整路径
+import { ChunkType } from '@renderer/types' // 调整路径
+
+export const createSimpleLoggingMiddleware = (): CompletionsMiddleware => {
+  return (api: MiddlewareAPI<AiProviderMiddlewareCompletionsContext, [CompletionsParams]>) => {
+    // console.log(`[LoggingMiddleware] Initialized for provider: ${api.getProviderId()}`);
+
+    return (next: (context: AiProviderMiddlewareCompletionsContext, params: CompletionsParams) => Promise<any>) => {
+      return async (context: AiProviderMiddlewareCompletionsContext, params: CompletionsParams): Promise<void> => {
+        const startTime = Date.now()
+        // 从 context 中获取 onChunk (它最初来自 params.onChunk)
+        const onChunk = context.onChunk
+
+        console.log(
+          `[LoggingMiddleware] Request for ${context.methodName} with params:`,
+          params.messages?.[params.messages.length - 1]?.content
+        )
+
+        try {
+          // 调用下一个中间件或核心逻辑
+          // `rawSdkResponse` 是来自下游的原始响应 (例如 OpenAIStream 或 ChatCompletion 对象)
+          const rawSdkResponse = await next(context, params)
+
+          // 此处简单示例不处理 rawSdkResponse，假设下游中间件 (如 StreamingResponseHandler)
+          // 会处理它并通过 onChunk 发送数据。
+          // 如果这个日志中间件在 StreamingResponseHandler 之后，那么流已经被处理。
+          // 如果在之前，那么它需要自己处理 rawSdkResponse 或确保下游会处理。
+
+          const duration = Date.now() - startTime
+          console.log(`[LoggingMiddleware] Request for ${context.methodName} completed in ${duration}ms.`)
+
+          // 假设下游已经通过 onChunk 发送了所有数据。
+          // 如果这个中间件是链的末端，并且需要确保 BLOCK_COMPLETE 被发送，
+          // 它可能需要更复杂的逻辑来跟踪何时所有数据都已发送。
+        } catch (error) {
+          const duration = Date.now() - startTime
+          console.error(`[LoggingMiddleware] Request for ${context.methodName} failed after ${duration}ms:`, error)
+
+          // 如果 onChunk 可用，可以尝试发送一个错误块
+          if (onChunk) {
+            onChunk({
+              type: ChunkType.ERROR,
+              error: { message: (error as Error).message, name: (error as Error).name, stack: (error as Error).stack }
+            })
+            // 考虑是否还需要发送 BLOCK_COMPLETE 来结束流
+            onChunk({ type: ChunkType.BLOCK_COMPLETE, response: {} })
+          }
+          throw error // 重新抛出错误，以便上层或全局错误处理器可以捕获
+        }
+      }
+    }
+  }
+}
+```
+
+### `AiProviderMiddlewareCompletionsContext` 的重要性
+
+`AiProviderMiddlewareCompletionsContext` 是在中间件之间传递状态和数据的核心。它通常包含：
+
+- `methodName`: 当前调用的方法名 (总是 `'completions'`)。
+- `originalArgs`: 传递给 `completions` 的原始参数数组。
+- `providerId`: Provider 的 ID。
+- `_providerInstance`: Provider 实例。
+- `onChunk`: 从原始 `CompletionsParams` 传入的回调函数，用于流式发送数据块。**所有中间件都应该通过 `context.onChunk` 来发送数据。**
+- `messages`, `model`, `assistant`, `mcpTools`: 从原始 `CompletionsParams` 中提取的常用字段，方便访问。
+- **自定义字段**: 中间件可以向上下文中添加自定义字段，以供后续中间件使用。例如，一个缓存中间件可能会添加 `context.cacheHit = true`。
+
+**关键**: 当你在中间件中修改 `params` 或 `context` 时，这些修改会向下游中间件传播（如果它们在 `next` 调用之前修改）。
+
+### 中间件的顺序
+
+中间件的执行顺序非常重要。它们在 `AiProviderMiddlewareConfig` 的数组中定义的顺序就是它们的执行顺序。
+
+- 请求首先通过第一个中间件，然后是第二个，依此类推。
+- 响应（或 `next` 的调用结果）则以相反的顺序"冒泡"回来。
+
+例如，如果链是 `[AuthMiddleware, CacheMiddleware, LoggingMiddleware]`：
+
+1.  `AuthMiddleware` 先执行其 "调用 `next` 之前" 的逻辑。
+2.  然后 `CacheMiddleware` 执行其 "调用 `next` 之前" 的逻辑。
+3.  然后 `LoggingMiddleware` 执行其 "调用 `next` 之前" 的逻辑。
+4.  核心SDK调用（或链的末端）。
+5.  `LoggingMiddleware` 先接收到结果，执行其 "调用 `next` 之后" 的逻辑。
+6.  然后 `CacheMiddleware` 接收到结果（可能已被 LoggingMiddleware 修改的上下文），执行其 "调用 `next` 之后" 的逻辑（例如，存储结果）。
+7.  最后 `AuthMiddleware` 接收到结果，执行其 "调用 `next` 之后" 的逻辑。
+
+### 注册中间件
+
+中间件在 `src/renderer/src/providers/middleware/register.ts` (或其他类似的配置文件) 中进行注册。
+
+```typescript
+// register.ts
+import { AiProviderMiddlewareConfig } from './AiProviderMiddlewareTypes'
+import { createSimpleLoggingMiddleware } from './common/SimpleLoggingMiddleware' // 假设你创建了这个文件
+import { createCompletionsLoggingMiddleware } from './common/CompletionsLoggingMiddleware' // 已有的
+
+const middlewareConfig: AiProviderMiddlewareConfig = {
+  completions: [
+    createSimpleLoggingMiddleware(), // 你新加的中间件
+    createCompletionsLoggingMiddleware() // 已有的日志中间件
+    // ... 其他 completions 中间件
+  ],
+  methods: {
+    // translate: [createGenericLoggingMiddleware()],
+    // ... 其他方法的中间件
+  }
+}
+
+export default middlewareConfig
+```
+
+### 最佳实践
+
+1.  **单一职责**: 每个中间件应专注于一个特定的功能（例如，日志、缓存、转换特定数据）。
+2.  **无副作用 (尽可能)**: 除了通过 `context` 或 `onChunk` 明确的副作用外，尽量避免修改全局状态或产生其他隐蔽的副作用。
+3.  **错误处理**:
+    - 在中间件内部使用 `try...catch` 来处理可能发生的错误。
+    - 决定是自行处理错误（例如，通过 `onChunk` 发送错误块）还是将错误重新抛出给上游。
+    - 如果重新抛出，确保错误对象包含足够的信息。
+4.  **性能考虑**: 中间件会增加请求处理的开销。避免在中间件中执行非常耗时的同步操作。对于IO密集型操作，确保它们是异步的。
+5.  **可配置性**: 使中间件的行为可通过参数或配置进行调整。例如，日志中间件可以接受一个日志级别参数。
+6.  **上下文管理**:
+    - 谨慎地向 `context` 添加数据。避免污染 `context` 或添加过大的对象。
+    - 明确你添加到 `context` 的字段的用途和生命周期。
+7.  **`next` 的调用**:
+    - 除非你有充分的理由提前终止请求（例如，缓存命中、授权失败），否则**总是确保调用 `await next(context, params)`**。否则，下游的中间件和核心逻辑将不会执行。
+    - 理解 `next` 的返回值并正确处理它，特别是当它是一个流时。你需要负责消费这个流或将其传递给另一个能够消费它的组件/中间件。
+8.  **命名清晰**: 给你的中间件和它们创建的函数起描述性的名字。
+9.  **文档和注释**: 对复杂的中间件逻辑添加注释，解释其工作原理和目的。
+
+### 调试技巧
+
+- 在中间件的关键点使用 `console.log` 或调试器来检查 `params`、`context` 的状态以及 `next` 的返回值。
+- 暂时简化中间件链，只保留你正在调试的中间件和最简单的核心逻辑，以隔离问题。
+- 编写单元测试来独立验证每个中间件的行为。
+
+通过遵循这些指南，你应该能够有效地为我们的系统创建强大且可维护的中间件。如果你有任何疑问或需要进一步的帮助，请咨询团队。

electron-builder.yml

@@ -11,6 +11,11 @@ electronLanguages:
   - en # for macOS
 directories:
   buildResources: build
+
+protocols:
+  - name: Cherry Studio
+    schemes:
+      - cherrystudio
 files:
   - '**/*'
   - '!**/{.vscode,.yarn,.yarn-lock,.github,.cursorrules,.prettierrc}'
@@ -48,7 +53,11 @@ files:
   - '!node_modules/pdf-parse/lib/pdf.js/{v1.9.426,v1.10.88,v2.0.550}'
   - '!node_modules/mammoth/{mammoth.browser.js,mammoth.browser.min.js}'
   - '!node_modules/selection-hook/prebuilds/**/*' # we rebuild .node, don't use prebuilds
-  - '!**/*.{h,iobj,ipdb,tlog,recipe,vcxproj,vcxproj.filters}' # filter .node build files
+  - '!node_modules/pdfjs-dist/web/**/*'
+  - '!node_modules/pdfjs-dist/legacy/web/*'
+  - '!node_modules/selection-hook/node_modules' # we don't need what in the node_modules dir
+  - '!node_modules/selection-hook/src' # we don't need source files
+  - '!**/*.{h,iobj,ipdb,tlog,recipe,vcxproj,vcxproj.filters,Makefile,*.Makefile}' # filter .node build files
 asarUnpack:
   - resources/**
   - '**/*.{metal,exp,lib}'
@@ -90,6 +99,7 @@ linux:
   artifactName: ${productName}-${version}-${arch}.${ext}
   target:
     - target: AppImage
+    - target: deb
   maintainer: electronjs.org
   category: Utility
   desktop:
@@ -107,8 +117,9 @@ afterSign: scripts/notarize.js
 artifactBuildCompleted: scripts/artifact-build-completed.js
 releaseInfo:
   releaseNotes: |
-    新增划词助手
-    助手支持分组
-    支持主题颜色切换
-    划词助手支持应用过滤
-    翻译模块功能改进
+    划词助手：支持 macOS 系统
+    文档处理：增加 MinerU、Doc2x，Mistral 等服务商支持
+    知识库：新的知识库界面，增加扫描版 PDF 支持
+    OCR：macOS 增加系统 OCR 支持
+    服务商：支持一键添加服务商，新增 PH8 大模型开放平台, 支持 PPIO OAuth 登录
+    修复：Linux下数据目录移动问题

electron.vite.config.ts

@@ -1,4 +1,5 @@
 import react from '@vitejs/plugin-react-swc'
+import { CodeInspectorPlugin } from 'code-inspector-plugin'
 import { defineConfig, externalizeDepsPlugin } from 'electron-vite'
 import { resolve } from 'path'
 import { visualizer } from 'rollup-plugin-visualizer'
@@ -19,7 +20,13 @@ export default defineConfig({
     },
     build: {
       rollupOptions: {
-        external: ['@libsql/client', 'bufferutil', 'utf-8-validate']
+        external: ['@libsql/client', 'bufferutil', 'utf-8-validate', '@cherrystudio/mac-system-ocr'],
+        output: {
+          // 彻底禁用代码分割 - 返回 null 强制单文件打包
+          manualChunks: undefined,
+          // 内联所有动态导入，这是关键配置
+          inlineDynamicImports: true
+        }
       },
       sourcemap: process.env.NODE_ENV === 'development'
     },
@@ -40,7 +47,6 @@ export default defineConfig({
   },
   renderer: {
     plugins: [
-      (async () => (await import('@tailwindcss/vite')).default())(),
       react({
         plugins: [
           [
@@ -54,6 +60,14 @@ export default defineConfig({
           ]
         ]
       }),
+      // 只在开发环境下启用 CodeInspectorPlugin
+      ...(process.env.NODE_ENV === 'development'
+        ? [
+            CodeInspectorPlugin({
+              bundler: 'vite'
+            })
+          ]
+        : []),
       ...visualizerPlugin('renderer')
     ],
     resolve: {
@@ -63,12 +77,16 @@ export default defineConfig({
       }
     },
     optimizeDeps: {
-      exclude: ['pyodide']
+      exclude: ['pyodide'],
+      esbuildOptions: {
+        target: 'esnext' // for dev
+      }
     },
     worker: {
       format: 'es'
     },
     build: {
+      target: 'esnext', // for build
       rollupOptions: {
         input: {
           index: resolve(__dirname, 'src/renderer/index.html'),

eslint.config.mjs

@@ -62,8 +62,7 @@ export default defineConfig([
       '.yarn/**',
       '.gitignore',
       'scripts/cloudflare-worker.js',
-      'src/main/integration/nutstore/sso/lib/**',
-      'src/renderer/src/ui/**'
+      'src/main/integration/nutstore/sso/lib/**'
     ]
   }
 ])

package.json

@@ -1,6 +1,6 @@
 {
   "name": "CherryStudio",
-  "version": "1.4.1",
+  "version": "1.4.8",
   "private": true,
   "description": "A powerful AI assistant for producer.",
   "main": "./out/main/index.js",
@@ -58,6 +58,25 @@
     "prepare": "husky"
   },
   "dependencies": {
+    "@cherrystudio/pdf-to-img-napi": "^0.0.1",
+    "@libsql/client": "0.14.0",
+    "@libsql/win32-x64-msvc": "^0.4.7",
+    "@strongtz/win32-arm64-msvc": "^0.4.7",
+    "jsdom": "26.1.0",
+    "macos-release": "^3.4.0",
+    "node-stream-zip": "^1.15.0",
+    "notion-helper": "^1.3.22",
+    "os-proxy-config": "^1.1.2",
+    "pdfjs-dist": "4.10.38",
+    "selection-hook": "^1.0.4",
+    "turndown": "7.2.0"
+  },
+  "devDependencies": {
+    "@agentic/exa": "^7.3.3",
+    "@agentic/searxng": "^7.3.3",
+    "@agentic/tavily": "^7.3.3",
+    "@ant-design/v5-patch-for-react-19": "^1.0.3",
+    "@anthropic-ai/sdk": "^0.41.0",
     "@cherrystudio/embedjs": "^0.1.31",
     "@cherrystudio/embedjs-libsql": "^0.1.31",
     "@cherrystudio/embedjs-loader-csv": "^0.1.31",
@@ -68,69 +87,31 @@
     "@cherrystudio/embedjs-loader-sitemap": "^0.1.31",
     "@cherrystudio/embedjs-loader-web": "^0.1.31",
     "@cherrystudio/embedjs-loader-xml": "^0.1.31",
+    "@cherrystudio/embedjs-ollama": "^0.1.31",
     "@cherrystudio/embedjs-openai": "^0.1.31",
-    "@electron-toolkit/utils": "^3.0.0",
-    "@langchain/community": "^0.3.36",
-    "@strongtz/win32-arm64-msvc": "^0.4.7",
-    "@tanstack/react-query": "^5.27.0",
-    "@types/react-infinite-scroll-component": "^5.0.0",
-    "archiver": "^7.0.1",
-    "async-mutex": "^0.5.0",
-    "diff": "^7.0.0",
-    "docx": "^9.0.2",
-    "electron-log": "^5.1.5",
-    "electron-store": "^8.2.0",
-    "electron-updater": "6.6.4",
-    "electron-window-state": "^5.0.3",
-    "epub": "patch:epub@npm%3A1.3.0#~/.yarn/patches/epub-npm-1.3.0-8325494ffe.patch",
-    "fast-xml-parser": "^5.2.0",
-    "franc-min": "^6.2.0",
-    "fs-extra": "^11.2.0",
-    "jsdom": "^26.0.0",
-    "markdown-it": "^14.1.0",
-    "node-stream-zip": "^1.15.0",
-    "officeparser": "^4.1.1",
-    "os-proxy-config": "^1.1.2",
-    "proxy-agent": "^6.5.0",
-    "selection-hook": "^0.9.22",
-    "tar": "^7.4.3",
-    "turndown": "^7.2.0",
-    "webdav": "^5.8.0",
-    "zipread": "^1.3.3"
-  },
-  "devDependencies": {
-    "@agentic/exa": "^7.3.3",
-    "@agentic/searxng": "^7.3.3",
-    "@agentic/tavily": "^7.3.3",
-    "@ant-design/v5-patch-for-react-19": "^1.0.3",
-    "@anthropic-ai/sdk": "^0.41.0",
     "@electron-toolkit/eslint-config-prettier": "^3.0.0",
     "@electron-toolkit/eslint-config-ts": "^3.0.0",
     "@electron-toolkit/preload": "^3.0.0",
     "@electron-toolkit/tsconfig": "^1.0.1",
+    "@electron-toolkit/utils": "^3.0.0",
     "@electron/notarize": "^2.5.0",
     "@emotion/is-prop-valid": "^1.3.1",
     "@eslint-react/eslint-plugin": "^1.36.1",
     "@eslint/js": "^9.22.0",
-    "@google/genai": "^1.0.1",
+    "@google/genai": "patch:@google/genai@npm%3A1.0.1#~/.yarn/patches/@google-genai-npm-1.0.1-e26f0f9af7.patch",
     "@hello-pangea/dnd": "^16.6.0",
     "@kangfenmao/keyv-storage": "^0.1.0",
+    "@langchain/community": "^0.3.36",
+    "@langchain/ollama": "^0.2.1",
+    "@mistralai/mistralai": "^1.6.0",
     "@modelcontextprotocol/sdk": "^1.11.4",
     "@mozilla/readability": "^0.6.0",
     "@notionhq/client": "^2.2.15",
     "@playwright/test": "^1.52.0",
-    "@radix-ui/react-collapsible": "^1.1.10",
-    "@radix-ui/react-dialog": "^1.1.13",
-    "@radix-ui/react-dropdown-menu": "^2.1.14",
-    "@radix-ui/react-separator": "^1.1.6",
-    "@radix-ui/react-slot": "^1.2.2",
-    "@radix-ui/react-tabs": "^1.1.11",
-    "@radix-ui/react-tooltip": "^1.2.6",
     "@reduxjs/toolkit": "^2.2.5",
-    "@shikijs/markdown-it": "^3.4.2",
+    "@shikijs/markdown-it": "^3.7.0",
     "@swc/plugin-styled-components": "^7.1.5",
-    "@tailwindcss/vite": "^4.1.5",
-    "@tavily/core": "patch:@tavily/core@npm%3A0.3.1#~/.yarn/patches/@tavily-core-npm-0.3.1-fe69bf2bea.patch",
+    "@tanstack/react-query": "^5.27.0",
     "@testing-library/dom": "^10.4.0",
     "@testing-library/jest-dom": "^6.6.3",
     "@testing-library/react": "^16.3.0",
@@ -147,36 +128,50 @@
     "@types/react-infinite-scroll-component": "^5.0.0",
     "@types/react-window": "^1",
     "@types/tinycolor2": "^1",
-    "@uiw/codemirror-extensions-langs": "^4.23.12",
-    "@uiw/codemirror-themes-all": "^4.23.12",
-    "@uiw/react-codemirror": "^4.23.12",
+    "@types/word-extractor": "^1",
+    "@uiw/codemirror-extensions-langs": "^4.23.14",
+    "@uiw/codemirror-themes-all": "^4.23.14",
+    "@uiw/react-codemirror": "^4.23.14",
     "@vitejs/plugin-react-swc": "^3.9.0",
     "@vitest/browser": "^3.1.4",
     "@vitest/coverage-v8": "^3.1.4",
     "@vitest/ui": "^3.1.4",
     "@vitest/web-worker": "^3.1.4",
     "@xyflow/react": "^12.4.4",
-    "antd": "^5.22.5",
+    "antd": "patch:antd@npm%3A5.24.7#~/.yarn/patches/antd-npm-5.24.7-356a553ae5.patch",
+    "archiver": "^7.0.1",
+    "async-mutex": "^0.5.0",
     "axios": "^1.7.3",
     "browser-image-compression": "^2.0.2",
-    "class-variance-authority": "^0.7.1",
-    "clsx": "^2.1.1",
+    "code-inspector-plugin": "^0.20.14",
     "color": "^5.0.0",
+    "country-flag-emoji-polyfill": "0.1.8",
     "dayjs": "^1.11.11",
     "dexie": "^4.0.8",
     "dexie-react-hooks": "^1.1.7",
+    "diff": "^7.0.0",
+    "docx": "^9.0.2",
     "dotenv-cli": "^7.4.2",
-    "electron": "35.4.0",
+    "electron": "35.6.0",
     "electron-builder": "26.0.15",
     "electron-devtools-installer": "^3.2.0",
+    "electron-log": "^5.1.5",
+    "electron-store": "^8.2.0",
+    "electron-updater": "6.6.4",
     "electron-vite": "^3.1.0",
+    "electron-window-state": "^5.0.3",
     "emittery": "^1.0.3",
     "emoji-picker-element": "^1.22.1",
+    "epub": "patch:epub@npm%3A1.3.0#~/.yarn/patches/epub-npm-1.3.0-8325494ffe.patch",
     "eslint": "^9.22.0",
     "eslint-plugin-react-hooks": "^5.2.0",
     "eslint-plugin-simple-import-sort": "^12.1.1",
     "eslint-plugin-unused-imports": "^4.1.4",
     "fast-diff": "^1.3.0",
+    "fast-xml-parser": "^5.2.0",
+    "franc-min": "^6.2.0",
+    "fs-extra": "^11.2.0",
+    "google-auth-library": "^9.15.1",
     "html-to-image": "^1.11.13",
     "husky": "^9.1.7",
     "i18next": "^23.11.5",
@@ -184,24 +179,25 @@
     "lint-staged": "^15.5.0",
     "lodash": "^4.17.21",
     "lru-cache": "^11.1.0",
-    "lucide-react": "^0.511.0",
-    "mermaid": "^11.6.0",
+    "lucide-react": "^0.487.0",
+    "markdown-it": "^14.1.0",
+    "mermaid": "^11.7.0",
     "mime": "^4.0.4",
-    "motion": "^12.12.1",
-    "next-themes": "^0.4.6",
+    "motion": "^12.10.5",
     "npx-scope-finder": "^1.2.0",
+    "officeparser": "^4.1.1",
     "openai": "patch:openai@npm%3A5.1.0#~/.yarn/patches/openai-npm-5.1.0-0e7b3ccb07.patch",
     "p-queue": "^8.1.0",
     "playwright": "^1.52.0",
     "prettier": "^3.5.3",
-    "prettier-plugin-tailwindcss": "^0.6.11",
+    "proxy-agent": "^6.5.0",
     "rc-virtual-list": "^3.18.6",
     "react": "^19.0.0",
     "react-dom": "^19.0.0",
     "react-hotkeys-hook": "^4.6.1",
     "react-i18next": "^14.1.2",
     "react-infinite-scroll-component": "^6.1.0",
-    "react-markdown": "^9.0.1",
+    "react-markdown": "^10.1.0",
     "react-redux": "^9.1.2",
     "react-router": "6",
     "react-router-dom": "6",
@@ -210,27 +206,30 @@
     "redux": "^5.0.1",
     "redux-persist": "^6.0.0",
     "rehype-katex": "^7.0.1",
-    "rehype-mathjax": "^7.0.0",
+    "rehype-mathjax": "^7.1.0",
     "rehype-raw": "^7.0.0",
-    "remark-cjk-friendly": "^1.1.0",
-    "remark-gfm": "^4.0.0",
+    "remark-cjk-friendly": "^1.2.0",
+    "remark-gfm": "^4.0.1",
     "remark-math": "^6.0.0",
+    "remove-markdown": "^0.6.2",
     "rollup-plugin-visualizer": "^5.12.0",
     "sass": "^1.88.0",
-    "shiki": "^3.4.2",
-    "sonner": "^2.0.3",
+    "shiki": "^3.7.0",
     "string-width": "^7.2.0",
     "styled-components": "^6.1.11",
-    "tailwind-merge": "^3.2.0",
-    "tailwindcss": "^4.1.5",
+    "tar": "^7.4.3",
     "tiny-pinyin": "^1.3.2",
-    "tokenx": "^0.4.1",
-    "tw-animate-css": "^1.2.9",
+    "tokenx": "^1.1.0",
     "typescript": "^5.6.2",
-    "usehooks-ts": "^3.1.1",
     "uuid": "^10.0.0",
     "vite": "6.2.6",
-    "vitest": "^3.1.4"
+    "vitest": "^3.1.4",
+    "webdav": "^5.8.0",
+    "word-extractor": "^1.0.4",
+    "zipread": "^1.3.3"
+  },
+  "optionalDependencies": {
+    "@cherrystudio/mac-system-ocr": "^0.2.2"
   },
   "resolutions": {
     "pdf-parse@npm:1.1.1": "patch:pdf-parse@npm%3A1.1.1#~/.yarn/patches/pdf-parse-npm-1.1.1-04a6109b2a.patch",

packages/shared/IpcChannel.ts

@@ -3,6 +3,8 @@ export enum IpcChannel {
   App_ClearCache = 'app:clear-cache',
   App_SetLaunchOnBoot = 'app:set-launch-on-boot',
   App_SetLanguage = 'app:set-language',
+  App_SetEnableSpellCheck = 'app:set-enable-spell-check',
+  App_SetSpellCheckLanguages = 'app:set-spell-check-languages',
   App_ShowUpdateDialog = 'app:show-update-dialog',
   App_CheckForUpdate = 'app:check-for-update',
   App_Reload = 'app:reload',
@@ -13,20 +15,33 @@ export enum IpcChannel {
   App_SetTrayOnClose = 'app:set-tray-on-close',
   App_SetTheme = 'app:set-theme',
   App_SetAutoUpdate = 'app:set-auto-update',
-  App_SetFeedUrl = 'app:set-feed-url',
+  App_SetTestPlan = 'app:set-test-plan',
+  App_SetTestChannel = 'app:set-test-channel',
   App_HandleZoomFactor = 'app:handle-zoom-factor',
-
+  App_Select = 'app:select',
+  App_HasWritePermission = 'app:has-write-permission',
+  App_Copy = 'app:copy',
+  App_SetStopQuitApp = 'app:set-stop-quit-app',
+  App_SetAppDataPath = 'app:set-app-data-path',
+  App_GetDataPathFromArgs = 'app:get-data-path-from-args',
+  App_FlushAppData = 'app:flush-app-data',
+  App_IsNotEmptyDir = 'app:is-not-empty-dir',
+  App_RelaunchApp = 'app:relaunch-app',
   App_IsBinaryExist = 'app:is-binary-exist',
   App_GetBinaryPath = 'app:get-binary-path',
   App_InstallUvBinary = 'app:install-uv-binary',
   App_InstallBunBinary = 'app:install-bun-binary',
 
+  App_MacIsProcessTrusted = 'app:mac-is-process-trusted',
+  App_MacRequestProcessTrust = 'app:mac-request-process-trust',
+
   App_QuoteToMain = 'app:quote-to-main',
 
   Notification_Send = 'notification:send',
   Notification_OnClick = 'notification:on-click',
 
   Webview_SetOpenLinkExternal = 'webview:set-open-link-external',
+  Webview_SetSpellCheckEnabled = 'webview:set-spell-check-enabled',
 
   // Open
   Open_Path = 'open:path',
@@ -59,6 +74,9 @@ export enum IpcChannel {
   Mcp_ServersUpdated = 'mcp:servers-updated',
   Mcp_CheckConnectivity = 'mcp:check-connectivity',
 
+  // Python
+  Python_Execute = 'python:execute',
+
   //copilot
   Copilot_GetAuthMessage = 'copilot:get-auth-message',
   Copilot_GetCopilotToken = 'copilot:get-copilot-token',
@@ -86,6 +104,10 @@ export enum IpcChannel {
   Gemini_ListFiles = 'gemini:list-files',
   Gemini_DeleteFile = 'gemini:delete-file',
 
+  // VertexAI
+  VertexAI_GetAuthHeaders = 'vertexai:get-auth-headers',
+  VertexAI_ClearAuthCache = 'vertexai:clear-auth-cache',
+
   Windows_ResetMinimumSize = 'window:reset-minimum-size',
   Windows_SetMinimumSize = 'window:set-minimum-size',
 
@@ -96,6 +118,7 @@ export enum IpcChannel {
   KnowledgeBase_Remove = 'knowledge-base:remove',
   KnowledgeBase_Search = 'knowledge-base:search',
   KnowledgeBase_Rerank = 'knowledge-base:rerank',
+  KnowledgeBase_Check_Quota = 'knowledge-base:check-quota',
 
   //file
   File_Open = 'file:open',
@@ -106,9 +129,10 @@ export enum IpcChannel {
   File_Clear = 'file:clear',
   File_Read = 'file:read',
   File_Delete = 'file:delete',
+  File_DeleteDir = 'file:deleteDir',
   File_Get = 'file:get',
   File_SelectFolder = 'file:selectFolder',
-  File_Create = 'file:create',
+  File_CreateTempFile = 'file:createTempFile',
   File_Write = 'file:write',
   File_WriteWithId = 'file:writeWithId',
   File_SaveImage = 'file:saveImage',
@@ -118,8 +142,15 @@ export enum IpcChannel {
   File_Copy = 'file:copy',
   File_BinaryImage = 'file:binaryImage',
   File_Base64File = 'file:base64File',
+  File_GetPdfInfo = 'file:getPdfInfo',
   Fs_Read = 'fs:read',
 
+  // file service
+  FileService_Upload = 'file-service:upload',
+  FileService_List = 'file-service:list',
+  FileService_Delete = 'file-service:delete',
+  FileService_Retrieve = 'file-service:retrieve',
+
   Export_Word = 'export:word',
 
   Shortcuts_Update = 'shortcuts:update',

packages/shared/config/constant.ts

@@ -1,7 +1,7 @@
 export const imageExts = ['.jpg', '.jpeg', '.png', '.gif', '.bmp', '.webp']
 export const videoExts = ['.mp4', '.avi', '.mov', '.wmv', '.flv', '.mkv']
 export const audioExts = ['.mp3', '.wav', '.ogg', '.flac', '.aac']
-export const documentExts = ['.pdf', '.docx', '.pptx', '.xlsx', '.odt', '.odp', '.ods']
+export const documentExts = ['.pdf', '.doc', '.docx', '.pptx', '.xlsx', '.odt', '.odp', '.ods']
 export const thirdPartyApplicationExts = ['.draftsExport']
 export const bookExts = ['.epub']
 const textExtsByCategory = new Map([
@@ -406,5 +406,16 @@ export const defaultLanguage = 'en-US'
 
 export enum FeedUrl {
   PRODUCTION = 'https://releases.cherry-ai.com',
-  EARLY_ACCESS = 'https://github.com/CherryHQ/cherry-studio/releases/latest/download'
+  GITHUB_LATEST = 'https://github.com/CherryHQ/cherry-studio/releases/latest/download',
+  PRERELEASE_LOWEST = 'https://github.com/CherryHQ/cherry-studio/releases/download/v1.4.0'
 }
+
+export enum UpgradeChannel {
+  LATEST = 'latest', // 最新稳定版本
+  RC = 'rc', // 公测版本
+  BETA = 'beta' // 预览版本
+}
+
+export const defaultTimeout = 10 * 1000 * 60
+
+export const occupiedDirs = ['logs', 'Network', 'Partitions/webview/Network']

packages/shared/config/types.ts

@@ -1,6 +1,11 @@
+import { ProcessingStatus } from '@types'
+
 export type LoaderReturn = {
   entriesAdded: number
   uniqueId: string
   uniqueIds: string[]
   loaderType: string
+  status?: ProcessingStatus
+  message?: string
+  messageSource?: 'preprocess' | 'embedding'
 }

resources/scripts/install-bun.js

@@ -2,12 +2,12 @@ const fs = require('fs')
 const path = require('path')
 const os = require('os')
 const { execSync } = require('child_process')
-const AdmZip = require('adm-zip')
+const StreamZip = require('node-stream-zip')
 const { downloadWithRedirects } = require('./download')
 
 // Base URL for downloading bun binaries
 const BUN_RELEASE_BASE_URL = 'https://gitcode.com/CherryHQ/bun/releases/download'
-const DEFAULT_BUN_VERSION = '1.2.9' // Default fallback version
+const DEFAULT_BUN_VERSION = '1.2.17' // Default fallback version
 
 // Mapping of platform+arch to binary package name
 const BUN_PACKAGES = {
@@ -66,35 +66,36 @@ async function downloadBunBinary(platform, arch, version = DEFAULT_BUN_VERSION,
 
     // Extract the zip file using adm-zip
     console.log(`Extracting ${packageName} to ${binDir}...`)
-    const zip = new AdmZip(tempFilename)
-    zip.extractAllTo(tempdir, true)
+    const zip = new StreamZip.async({ file: tempFilename })
 
-    // Move files using Node.js fs
-    const sourceDir = path.join(tempdir, packageName.split('.')[0])
-    const files = fs.readdirSync(sourceDir)
+    // Get all entries in the zip file
+    const entries = await zip.entries()
 
-    for (const file of files) {
-      const sourcePath = path.join(sourceDir, file)
-      const destPath = path.join(binDir, file)
+    // Extract files directly to binDir, flattening the directory structure
+    for (const entry of Object.values(entries)) {
+      if (!entry.isDirectory) {
+        // Get just the filename without path
+        const filename = path.basename(entry.name)
+        const outputPath = path.join(binDir, filename)
 
-      fs.copyFileSync(sourcePath, destPath)
-      fs.unlinkSync(sourcePath)
-
-      // Set executable permissions for non-Windows platforms
-      if (platform !== 'win32') {
-        try {
-          // 755 permission: rwxr-xr-x
-          fs.chmodSync(destPath, '755')
-        } catch (error) {
-          console.warn(`Warning: Failed to set executable permissions: ${error.message}`)
+        console.log(`Extracting ${entry.name} -> ${filename}`)
+        await zip.extract(entry.name, outputPath)
+        // Make executable files executable on Unix-like systems
+        if (platform !== 'win32') {
+          try {
+            fs.chmodSync(outputPath, 0o755)
+          } catch (chmodError) {
+            console.error(`Warning: Failed to set executable permissions on ${filename}`)
+            return false
+          }
         }
+        console.log(`Extracted ${entry.name} -> ${outputPath}`)
       }
     }
+    await zip.close()
 
     // Clean up
     fs.unlinkSync(tempFilename)
-    fs.rmSync(sourceDir, { recursive: true })
-
     console.log(`Successfully installed bun ${version} for ${platformKey}`)
     return true
   } catch (error) {

resources/scripts/install-uv.js

@@ -2,34 +2,33 @@ const fs = require('fs')
 const path = require('path')
 const os = require('os')
 const { execSync } = require('child_process')
-const tar = require('tar')
-const AdmZip = require('adm-zip')
+const StreamZip = require('node-stream-zip')
 const { downloadWithRedirects } = require('./download')
 
 // Base URL for downloading uv binaries
 const UV_RELEASE_BASE_URL = 'https://gitcode.com/CherryHQ/uv/releases/download'
-const DEFAULT_UV_VERSION = '0.6.14'
+const DEFAULT_UV_VERSION = '0.7.13'
 
 // Mapping of platform+arch to binary package name
 const UV_PACKAGES = {
-  'darwin-arm64': 'uv-aarch64-apple-darwin.tar.gz',
-  'darwin-x64': 'uv-x86_64-apple-darwin.tar.gz',
+  'darwin-arm64': 'uv-aarch64-apple-darwin.zip',
+  'darwin-x64': 'uv-x86_64-apple-darwin.zip',
   'win32-arm64': 'uv-aarch64-pc-windows-msvc.zip',
   'win32-ia32': 'uv-i686-pc-windows-msvc.zip',
   'win32-x64': 'uv-x86_64-pc-windows-msvc.zip',
-  'linux-arm64': 'uv-aarch64-unknown-linux-gnu.tar.gz',
-  'linux-ia32': 'uv-i686-unknown-linux-gnu.tar.gz',
-  'linux-ppc64': 'uv-powerpc64-unknown-linux-gnu.tar.gz',
-  'linux-ppc64le': 'uv-powerpc64le-unknown-linux-gnu.tar.gz',
-  'linux-s390x': 'uv-s390x-unknown-linux-gnu.tar.gz',
-  'linux-x64': 'uv-x86_64-unknown-linux-gnu.tar.gz',
-  'linux-armv7l': 'uv-armv7-unknown-linux-gnueabihf.tar.gz',
+  'linux-arm64': 'uv-aarch64-unknown-linux-gnu.zip',
+  'linux-ia32': 'uv-i686-unknown-linux-gnu.zip',
+  'linux-ppc64': 'uv-powerpc64-unknown-linux-gnu.zip',
+  'linux-ppc64le': 'uv-powerpc64le-unknown-linux-gnu.zip',
+  'linux-s390x': 'uv-s390x-unknown-linux-gnu.zip',
+  'linux-x64': 'uv-x86_64-unknown-linux-gnu.zip',
+  'linux-armv7l': 'uv-armv7-unknown-linux-gnueabihf.zip',
   // MUSL variants
-  'linux-musl-arm64': 'uv-aarch64-unknown-linux-musl.tar.gz',
-  'linux-musl-ia32': 'uv-i686-unknown-linux-musl.tar.gz',
-  'linux-musl-x64': 'uv-x86_64-unknown-linux-musl.tar.gz',
-  'linux-musl-armv6l': 'uv-arm-unknown-linux-musleabihf.tar.gz',
-  'linux-musl-armv7l': 'uv-armv7-unknown-linux-musleabihf.tar.gz'
+  'linux-musl-arm64': 'uv-aarch64-unknown-linux-musl.zip',
+  'linux-musl-ia32': 'uv-i686-unknown-linux-musl.zip',
+  'linux-musl-x64': 'uv-x86_64-unknown-linux-musl.zip',
+  'linux-musl-armv6l': 'uv-arm-unknown-linux-musleabihf.zip',
+  'linux-musl-armv7l': 'uv-armv7-unknown-linux-musleabihf.zip'
 }
 
 /**
@@ -66,46 +65,35 @@ async function downloadUvBinary(platform, arch, version = DEFAULT_UV_VERSION, is
 
     console.log(`Extracting ${packageName} to ${binDir}...`)
 
-    // 根据文件扩展名选择解压方法
-    if (packageName.endsWith('.zip')) {
-      // 使用 adm-zip 处理 zip 文件
-      const zip = new AdmZip(tempFilename)
-      zip.extractAllTo(binDir, true)
-      fs.unlinkSync(tempFilename)
-      console.log(`Successfully installed uv ${version} for ${platform}-${arch}`)
-      return true
-    } else {
-      // tar.gz 文件的处理保持不变
-      await tar.x({
-        file: tempFilename,
-        cwd: tempdir,
-        z: true
-      })
+    const zip = new StreamZip.async({ file: tempFilename })
 
-      // Move files using Node.js fs
-      const sourceDir = path.join(tempdir, packageName.split('.')[0])
-      const files = fs.readdirSync(sourceDir)
-      for (const file of files) {
-        const sourcePath = path.join(sourceDir, file)
-        const destPath = path.join(binDir, file)
-        fs.copyFileSync(sourcePath, destPath)
-        fs.unlinkSync(sourcePath)
+    // Get all entries in the zip file
+    const entries = await zip.entries()
 
-        // Set executable permissions for non-Windows platforms
+    // Extract files directly to binDir, flattening the directory structure
+    for (const entry of Object.values(entries)) {
+      if (!entry.isDirectory) {
+        // Get just the filename without path
+        const filename = path.basename(entry.name)
+        const outputPath = path.join(binDir, filename)
+
+        console.log(`Extracting ${entry.name} -> ${filename}`)
+        await zip.extract(entry.name, outputPath)
+        // Make executable files executable on Unix-like systems
         if (platform !== 'win32') {
           try {
-            fs.chmodSync(destPath, '755')
-          } catch (error) {
-            console.warn(`Warning: Failed to set executable permissions: ${error.message}`)
+            fs.chmodSync(outputPath, 0o755)
+          } catch (chmodError) {
+            console.error(`Warning: Failed to set executable permissions on ${filename}`)
+            return false
           }
         }
+        console.log(`Extracted ${entry.name} -> ${outputPath}`)
       }
-
-      // Clean up
-      fs.unlinkSync(tempFilename)
-      fs.rmSync(sourceDir, { recursive: true })
     }
 
+    await zip.close()
+    fs.unlinkSync(tempFilename)
     console.log(`Successfully installed uv ${version} for ${platform}-${arch}`)
     return true
   } catch (error) {

scripts/after-pack.js

@@ -23,6 +23,9 @@ exports.default = async function (context) {
     const node_modules_path = path.join(context.appOutDir, 'resources', 'app.asar.unpacked', 'node_modules')
     const _arch = arch === Arch.arm64 ? ['linux-arm64-gnu', 'linux-arm64-musl'] : ['linux-x64-gnu', 'linux-x64-musl']
     keepPackageNodeFiles(node_modules_path, '@libsql', _arch)
+
+    // 删除 macOS 专用的 OCR 包
+    removeMacOnlyPackages(node_modules_path)
   }
 
   if (platform === 'windows') {
@@ -35,7 +38,30 @@ exports.default = async function (context) {
       keepPackageNodeFiles(node_modules_path, '@strongtz', ['win32-x64-msvc'])
       keepPackageNodeFiles(node_modules_path, '@libsql', ['win32-x64-msvc'])
     }
+
+    removeMacOnlyPackages(node_modules_path)
   }
+
+  if (platform === 'windows') {
+    fs.rmSync(path.join(context.appOutDir, 'LICENSE.electron.txt'), { force: true })
+    fs.rmSync(path.join(context.appOutDir, 'LICENSES.chromium.html'), { force: true })
+  }
+}
+
+/**
+ * 删除 macOS 专用的包
+ * @param {string} nodeModulesPath
+ */
+function removeMacOnlyPackages(nodeModulesPath) {
+  const macOnlyPackages = ['@cherrystudio/mac-system-ocr']
+
+  macOnlyPackages.forEach((packageName) => {
+    const packagePath = path.join(nodeModulesPath, packageName)
+    if (fs.existsSync(packagePath)) {
+      fs.rmSync(packagePath, { recursive: true, force: true })
+      console.log(`[After Pack] Removed macOS-only package: ${packageName}`)
+    }
+  })
 }
 
 /**

scripts/update-i18n.ts

@@ -1,16 +1,19 @@
 /**
- * Paratera_API_KEY=sk-abcxxxxxxxxxxxxxxxxxxxxxxx123 ts-node scripts/update-i18n.ts
+ * 使用 OpenAI 兼容的模型生成 i18n 文本，并更新到 translate 目录
+ *
+ * API_KEY=sk-xxxx BASE_URL=xxxx MODEL=xxxx ts-node scripts/update-i18n.ts
  */
 
-// OCOOL API KEY
-const Paratera_API_KEY = process.env.Paratera_API_KEY
+const API_KEY = process.env.API_KEY
+const BASE_URL = process.env.BASE_URL || 'https://llmapi.paratera.com/v1'
+const MODEL = process.env.MODEL || 'Qwen3-235B-A22B'
 
 const INDEX = [
-  //         语言的名称          代码            用来翻译的模型
-  { name: 'France', code: 'fr-fr', model: 'Qwen3-235B-A22B' },
-  { name: 'Spanish', code: 'es-es', model: 'Qwen3-235B-A22B' },
-  { name: 'Portuguese', code: 'pt-pt', model: 'Qwen3-235B-A22B' },
-  { name: 'Greek', code: 'el-gr', model: 'Qwen3-235B-A22B' }
+  // 语言的名称代码用来翻译的模型
+  { name: 'France', code: 'fr-fr', model: MODEL },
+  { name: 'Spanish', code: 'es-es', model: MODEL },
+  { name: 'Portuguese', code: 'pt-pt', model: MODEL },
+  { name: 'Greek', code: 'el-gr', model: MODEL }
 ]
 
 const fs = require('fs')
@@ -19,8 +22,8 @@ import OpenAI from 'openai'
 const zh = JSON.parse(fs.readFileSync('src/renderer/src/i18n/locales/zh-cn.json', 'utf8')) as object
 
 const openai = new OpenAI({
-  apiKey: Paratera_API_KEY,
-  baseURL: 'https://llmapi.paratera.com/v1'
+  apiKey: API_KEY,
+  baseURL: BASE_URL
 })
 
 // 递归遍历翻译

src/main/bootstrap.ts

@@ -0,0 +1,33 @@
+import { occupiedDirs } from '@shared/config/constant'
+import { app } from 'electron'
+import fs from 'fs'
+import path from 'path'
+
+import { initAppDataDir } from './utils/file'
+
+app.isPackaged && initAppDataDir()
+
+// 在主进程中复制 appData 中某些一直被占用的文件
+// 在renderer进程还没有启动时，主进程可以复制这些文件到新的appData中
+function copyOccupiedDirsInMainProcess() {
+  const newAppDataPath = process.argv
+    .slice(1)
+    .find((arg) => arg.startsWith('--new-data-path='))
+    ?.split('--new-data-path=')[1]
+  if (!newAppDataPath) {
+    return
+  }
+
+  if (process.platform === 'win32') {
+    const appDataPath = app.getPath('userData')
+    occupiedDirs.forEach((dir) => {
+      const dirPath = path.join(appDataPath, dir)
+      const newDirPath = path.join(newAppDataPath, dir)
+      if (fs.existsSync(dirPath)) {
+        fs.cpSync(dirPath, newDirPath, { recursive: true })
+      }
+    })
+  }
+}
+
+copyOccupiedDirsInMainProcess()

src/main/config.ts

@@ -1,7 +1,6 @@
 import { app } from 'electron'
 
 import { getDataPath } from './utils'
-
 const isDev = process.env.NODE_ENV === 'development'
 
 if (isDev) {
@@ -11,13 +10,13 @@ if (isDev) {
 export const DATA_PATH = getDataPath()
 
 export const titleBarOverlayDark = {
-  height: 42,
+  height: 40,
   color: 'rgba(255,255,255,0)',
   symbolColor: '#fff'
 }
 
 export const titleBarOverlayLight = {
-  height: 42,
+  height: 40,
   color: 'rgba(255,255,255,0)',
   symbolColor: '#000'
 }

src/main/configs/SelectionConfig.ts

@@ -1,6 +1,6 @@
 interface IFilterList {
   WINDOWS: string[]
-  MAC?: string[]
+  MAC: string[]
 }
 
 interface IFinetunedList {
@@ -45,14 +45,17 @@ export const SELECTION_PREDEFINED_BLACKLIST: IFilterList = {
     'sldworks.exe',
     // Remote Desktop
     'mstsc.exe'
-  ]
+  ],
+  MAC: []
 }
 
 export const SELECTION_FINETUNED_LIST: IFinetunedList = {
   EXCLUDE_CLIPBOARD_CURSOR_DETECT: {
-    WINDOWS: ['acrobat.exe', 'wps.exe', 'cajviewer.exe']
+    WINDOWS: ['acrobat.exe', 'wps.exe', 'cajviewer.exe'],
+    MAC: []
   },
   INCLUDE_CLIPBOARD_DELAY_READ: {
-    WINDOWS: ['acrobat.exe', 'wps.exe', 'cajviewer.exe', 'foxitphantom.exe']
+    WINDOWS: ['acrobat.exe', 'wps.exe', 'cajviewer.exe', 'foxitphantom.exe'],
+    MAC: []
   }
 }

src/main/embeddings/EmbeddingsFactory.ts

@@ -1,38 +0,0 @@
-import type { BaseEmbeddings } from '@cherrystudio/embedjs-interfaces'
-import { OpenAiEmbeddings } from '@cherrystudio/embedjs-openai'
-import { AzureOpenAiEmbeddings } from '@cherrystudio/embedjs-openai/src/azure-openai-embeddings'
-import { getInstanceName } from '@main/utils'
-import { KnowledgeBaseParams } from '@types'
-
-import VoyageEmbeddings from './VoyageEmbeddings'
-
-export default class EmbeddingsFactory {
-  static create({ model, apiKey, apiVersion, baseURL, dimensions }: KnowledgeBaseParams): BaseEmbeddings {
-    const batchSize = 10
-    if (model.includes('voyage')) {
-      return new VoyageEmbeddings({
-        modelName: model,
-        apiKey,
-        outputDimension: dimensions,
-        batchSize: 8
-      })
-    }
-    if (apiVersion !== undefined) {
-      return new AzureOpenAiEmbeddings({
-        azureOpenAIApiKey: apiKey,
-        azureOpenAIApiVersion: apiVersion,
-        azureOpenAIApiDeploymentName: model,
-        azureOpenAIApiInstanceName: getInstanceName(baseURL),
-        dimensions,
-        batchSize
-      })
-    }
-    return new OpenAiEmbeddings({
-      model,
-      apiKey,
-      dimensions,
-      batchSize,
-      configuration: { baseURL }
-    })
-  }
-}

src/main/index.ts

@@ -1,3 +1,8 @@
+// don't reorder this file, it's used to initialize the app data dir and
+// other which should be run before the main process is ready
+// eslint-disable-next-line
+import './bootstrap'
+
 import '@main/config'
 
 import { electronApp, optimizer } from '@electron-toolkit/utils'
@@ -20,7 +25,6 @@ import selectionService, { initSelectionService } from './services/SelectionServ
 import { registerShortcuts } from './services/ShortcutService'
 import { TrayService } from './services/TrayService'
 import { windowService } from './services/WindowService'
-import { setUserDataDir } from './utils/file'
 
 Logger.initialize()
 
@@ -34,6 +38,26 @@ if (isWin) {
   app.commandLine.appendSwitch('wm-window-animations-disabled')
 }
 
+// Enable features for unresponsive renderer js call stacks
+app.commandLine.appendSwitch('enable-features', 'DocumentPolicyIncludeJSCallStacksInCrashReports')
+app.on('web-contents-created', (_, webContents) => {
+  webContents.session.webRequest.onHeadersReceived((details, callback) => {
+    callback({
+      responseHeaders: {
+        ...details.responseHeaders,
+        'Document-Policy': ['include-js-call-stacks-in-crash-reports']
+      }
+    })
+  })
+
+  webContents.on('unresponsive', async () => {
+    // Interrupt execution and collect call stack from unresponsive renderer
+    Logger.error('Renderer unresponsive start')
+    const callStack = await webContents.mainFrame.collectJavaScriptCallStack()
+    Logger.error('Renderer unresponsive js call stack\n', callStack)
+  })
+})
+
 // in production mode, handle uncaught exception and unhandled rejection globally
 if (!isDev) {
   // handle uncaught exception
@@ -52,9 +76,6 @@ if (!app.requestSingleInstanceLock()) {
   app.quit()
   process.exit(0)
 } else {
-  // Portable dir must be setup before app ready
-  setUserDataDir()
-
   // This method will be called when Electron has finished
   // initialization and is ready to create browser windows.
   // Some APIs can only be used after this event occurs.
@@ -103,19 +124,27 @@ if (!app.requestSingleInstanceLock()) {
   registerProtocolClient(app)
 
   // macOS specific: handle protocol when app is already running
+
   app.on('open-url', (event, url) => {
     event.preventDefault()
     handleProtocolUrl(url)
   })
 
+  const handleOpenUrl = (args: string[]) => {
+    const url = args.find((arg) => arg.startsWith(CHERRY_STUDIO_PROTOCOL + '://'))
+    if (url) handleProtocolUrl(url)
+  }
+
+  // for windows to start with url
+  handleOpenUrl(process.argv)
+
   // Listen for second instance
   app.on('second-instance', (_event, argv) => {
     windowService.showMainWindow()
 
     // Protocol handler for Windows/Linux
     // The commandLine is an array of strings where the last item might be the URL
-    const url = argv.find((arg) => arg.startsWith(CHERRY_STUDIO_PROTOCOL + '://'))
-    if (url) handleProtocolUrl(url)
+    handleOpenUrl(argv)
   })
 
   app.on('browser-window-created', (_, window) => {

src/main/ipc.ts

@@ -1,12 +1,14 @@
 import fs from 'node:fs'
 import { arch } from 'node:os'
+import path from 'node:path'
 
-import { isMac, isWin } from '@main/constant'
+import { isLinux, isMac, isWin } from '@main/constant'
 import { getBinaryPath, isBinaryExists, runInstallScript } from '@main/utils/process'
 import { handleZoomFactor } from '@main/utils/zoom'
+import { UpgradeChannel } from '@shared/config/constant'
 import { IpcChannel } from '@shared/IpcChannel'
-import { Shortcut, ThemeMode } from '@types'
-import { BrowserWindow, ipcMain, session, shell } from 'electron'
+import { FileMetadata, Provider, Shortcut, ThemeMode } from '@types'
+import { BrowserWindow, dialog, ipcMain, session, shell, systemPreferences, webContents } from 'electron'
 import log from 'electron-log'
 import { Notification } from 'src/renderer/src/types/notification'
 
@@ -15,36 +17,42 @@ import BackupManager from './services/BackupManager'
 import { configManager } from './services/ConfigManager'
 import CopilotService from './services/CopilotService'
 import { ExportService } from './services/ExportService'
-import FileService from './services/FileService'
 import FileStorage from './services/FileStorage'
+import FileService from './services/FileSystemService'
 import KnowledgeService from './services/KnowledgeService'
 import mcpService from './services/MCPService'
 import NotificationService from './services/NotificationService'
 import * as NutstoreService from './services/NutstoreService'
 import ObsidianVaultService from './services/ObsidianVaultService'
 import { ProxyConfig, proxyManager } from './services/ProxyManager'
+import { pythonService } from './services/PythonService'
+import { FileServiceManager } from './services/remotefile/FileServiceManager'
 import { searchService } from './services/SearchService'
 import { SelectionService } from './services/SelectionService'
 import { registerShortcuts, unregisterAllShortcuts } from './services/ShortcutService'
 import storeSyncService from './services/StoreSyncService'
 import { themeService } from './services/ThemeService'
+import VertexAIService from './services/VertexAIService'
 import { setOpenLinkExternal } from './services/WebviewService'
 import { windowService } from './services/WindowService'
 import { calculateDirectorySize, getResourcePath } from './utils'
 import { decrypt, encrypt } from './utils/aes'
-import { getCacheDir, getConfigDir, getFilesDir } from './utils/file'
+import { getCacheDir, getConfigDir, getFilesDir, hasWritePermission, updateAppDataConfig } from './utils/file'
 import { compress, decompress } from './utils/zip'
-import { FeedUrl } from '@shared/config/constant'
 
 const fileManager = new FileStorage()
 const backupManager = new BackupManager()
 const exportService = new ExportService(fileManager)
 const obsidianVaultService = new ObsidianVaultService()
+const vertexAIService = VertexAIService.getInstance()
 
 export function registerIpc(mainWindow: BrowserWindow, app: Electron.App) {
   const appUpdater = new AppUpdater(mainWindow)
   const notificationService = new NotificationService(mainWindow)
 
+  // Initialize Python service with main window
+  pythonService.setMainWindow(mainWindow)
+
   ipcMain.handle(IpcChannel.App_Info, () => ({
     version: app.getVersion(),
     isPackaged: app.isPackaged,
@@ -55,7 +63,8 @@ export function registerIpc(mainWindow: BrowserWindow, app: Electron.App) {
     resourcesPath: getResourcePath(),
     logsPath: log.transports.file.getFile().path,
     arch: arch(),
-    isPortable: isWin && 'PORTABLE_EXECUTABLE_DIR' in process.env
+    isPortable: isWin && 'PORTABLE_EXECUTABLE_DIR' in process.env,
+    installPath: path.dirname(app.getPath('exe'))
   }))
 
   ipcMain.handle(IpcChannel.App_Proxy, async (_, proxy: string) => {
@@ -83,6 +92,27 @@ export function registerIpc(mainWindow: BrowserWindow, app: Electron.App) {
     configManager.setLanguage(language)
   })
 
+  // spell check
+  ipcMain.handle(IpcChannel.App_SetEnableSpellCheck, (_, isEnable: boolean) => {
+    // disable spell check for all webviews
+    const webviews = webContents.getAllWebContents()
+    webviews.forEach((webview) => {
+      webview.session.setSpellCheckerEnabled(isEnable)
+    })
+  })
+
+  // spell check languages
+  ipcMain.handle(IpcChannel.App_SetSpellCheckLanguages, (_, languages: string[]) => {
+    if (languages.length === 0) {
+      return
+    }
+    const windows = BrowserWindow.getAllWindows()
+    windows.forEach((window) => {
+      window.webContents.session.setSpellCheckerLanguages(languages)
+    })
+    configManager.set('spellCheckLanguages', languages)
+  })
+
   // launch on boot
   ipcMain.handle(IpcChannel.App_SetLaunchOnBoot, (_, openAtLogin: boolean) => {
     // Set login item settings for windows and mac
@@ -113,10 +143,34 @@ export function registerIpc(mainWindow: BrowserWindow, app: Electron.App) {
     configManager.setAutoUpdate(isActive)
   })
 
-  ipcMain.handle(IpcChannel.App_SetFeedUrl, (_, feedUrl: FeedUrl) => {
-    appUpdater.setFeedUrl(feedUrl)
+  ipcMain.handle(IpcChannel.App_SetTestPlan, async (_, isActive: boolean) => {
+    log.info('set test plan', isActive)
+    if (isActive !== configManager.getTestPlan()) {
+      appUpdater.cancelDownload()
+      configManager.setTestPlan(isActive)
+    }
   })
 
+  ipcMain.handle(IpcChannel.App_SetTestChannel, async (_, channel: UpgradeChannel) => {
+    log.info('set test channel', channel)
+    if (channel !== configManager.getTestChannel()) {
+      appUpdater.cancelDownload()
+      configManager.setTestChannel(channel)
+    }
+  })
+
+  //only for mac
+  if (isMac) {
+    ipcMain.handle(IpcChannel.App_MacIsProcessTrusted, (): boolean => {
+      return systemPreferences.isTrustedAccessibilityClient(false)
+    })
+
+    //return is only the current state, not the new state
+    ipcMain.handle(IpcChannel.App_MacRequestProcessTrust, (): boolean => {
+      return systemPreferences.isTrustedAccessibilityClient(true)
+    })
+  }
+
   ipcMain.handle(IpcChannel.Config_Set, (_, key: string, value: any, isNotify: boolean = false) => {
     configManager.set(key, value, isNotify)
   })
@@ -173,6 +227,113 @@ export function registerIpc(mainWindow: BrowserWindow, app: Electron.App) {
     }
   })
 
+  let preventQuitListener: ((event: Electron.Event) => void) | null = null
+  ipcMain.handle(IpcChannel.App_SetStopQuitApp, (_, stop: boolean = false, reason: string = '') => {
+    if (stop) {
+      // Only add listener if not already added
+      if (!preventQuitListener) {
+        preventQuitListener = (event: Electron.Event) => {
+          event.preventDefault()
+          notificationService.sendNotification({
+            title: reason,
+            message: reason
+          } as Notification)
+        }
+        app.on('before-quit', preventQuitListener)
+      }
+    } else {
+      // Remove listener if it exists
+      if (preventQuitListener) {
+        app.removeListener('before-quit', preventQuitListener)
+        preventQuitListener = null
+      }
+    }
+  })
+
+  // Select app data path
+  ipcMain.handle(IpcChannel.App_Select, async (_, options: Electron.OpenDialogOptions) => {
+    try {
+      const { canceled, filePaths } = await dialog.showOpenDialog(options)
+      if (canceled || filePaths.length === 0) {
+        return null
+      }
+      return filePaths[0]
+    } catch (error: any) {
+      log.error('Failed to select app data path:', error)
+      return null
+    }
+  })
+
+  ipcMain.handle(IpcChannel.App_HasWritePermission, async (_, filePath: string) => {
+    return hasWritePermission(filePath)
+  })
+
+  // Set app data path
+  ipcMain.handle(IpcChannel.App_SetAppDataPath, async (_, filePath: string) => {
+    updateAppDataConfig(filePath)
+    app.setPath('userData', filePath)
+  })
+
+  ipcMain.handle(IpcChannel.App_GetDataPathFromArgs, () => {
+    return process.argv
+      .slice(1)
+      .find((arg) => arg.startsWith('--new-data-path='))
+      ?.split('--new-data-path=')[1]
+  })
+
+  ipcMain.handle(IpcChannel.App_FlushAppData, () => {
+    BrowserWindow.getAllWindows().forEach((w) => {
+      w.webContents.session.flushStorageData()
+      w.webContents.session.cookies.flushStore()
+
+      w.webContents.session.closeAllConnections()
+    })
+
+    session.defaultSession.flushStorageData()
+    session.defaultSession.cookies.flushStore()
+    session.defaultSession.closeAllConnections()
+  })
+
+  ipcMain.handle(IpcChannel.App_IsNotEmptyDir, async (_, path: string) => {
+    return fs.readdirSync(path).length > 0
+  })
+
+  // Copy user data to new location
+  ipcMain.handle(IpcChannel.App_Copy, async (_, oldPath: string, newPath: string, occupiedDirs: string[] = []) => {
+    try {
+      await fs.promises.cp(oldPath, newPath, {
+        recursive: true,
+        filter: (src) => {
+          if (occupiedDirs.some((dir) => src.startsWith(path.resolve(dir)))) {
+            return false
+          }
+          return true
+        }
+      })
+      return { success: true }
+    } catch (error: any) {
+      log.error('Failed to copy user data:', error)
+      return { success: false, error: error.message }
+    }
+  })
+
+  // Relaunch app
+  ipcMain.handle(IpcChannel.App_RelaunchApp, (_, options?: Electron.RelaunchOptions) => {
+    // Fix for .AppImage
+    if (isLinux && process.env.APPIMAGE) {
+      log.info('Relaunching app with options:', process.env.APPIMAGE, options)
+      // On Linux, we need to use the APPIMAGE environment variable to relaunch
+      // https://github.com/electron-userland/electron-builder/issues/1727#issuecomment-769896927
+      options = options || {}
+      options.execPath = process.env.APPIMAGE
+      options.args = options.args || []
+      options.args.unshift('--appimage-extract-and-run')
+    }
+
+    app.relaunch(options)
+    app.exit(0)
+  })
+
   // check for update
   ipcMain.handle(IpcChannel.App_CheckForUpdate, async () => {
     return await appUpdater.checkForUpdates()
@@ -217,19 +378,42 @@ export function registerIpc(mainWindow: BrowserWindow, app: Electron.App) {
   ipcMain.handle(IpcChannel.File_Clear, fileManager.clear)
   ipcMain.handle(IpcChannel.File_Read, fileManager.readFile)
   ipcMain.handle(IpcChannel.File_Delete, fileManager.deleteFile)
+  ipcMain.handle('file:deleteDir', fileManager.deleteDir)
   ipcMain.handle(IpcChannel.File_Get, fileManager.getFile)
   ipcMain.handle(IpcChannel.File_SelectFolder, fileManager.selectFolder)
-  ipcMain.handle(IpcChannel.File_Create, fileManager.createTempFile)
+  ipcMain.handle(IpcChannel.File_CreateTempFile, fileManager.createTempFile)
   ipcMain.handle(IpcChannel.File_Write, fileManager.writeFile)
   ipcMain.handle(IpcChannel.File_WriteWithId, fileManager.writeFileWithId)
   ipcMain.handle(IpcChannel.File_SaveImage, fileManager.saveImage)
   ipcMain.handle(IpcChannel.File_Base64Image, fileManager.base64Image)
   ipcMain.handle(IpcChannel.File_SaveBase64Image, fileManager.saveBase64Image)
   ipcMain.handle(IpcChannel.File_Base64File, fileManager.base64File)
+  ipcMain.handle(IpcChannel.File_GetPdfInfo, fileManager.pdfPageCount)
   ipcMain.handle(IpcChannel.File_Download, fileManager.downloadFile)
   ipcMain.handle(IpcChannel.File_Copy, fileManager.copyFile)
   ipcMain.handle(IpcChannel.File_BinaryImage, fileManager.binaryImage)
 
+  // file service
+  ipcMain.handle(IpcChannel.FileService_Upload, async (_, provider: Provider, file: FileMetadata) => {
+    const service = FileServiceManager.getInstance().getService(provider)
+    return await service.uploadFile(file)
+  })
+
+  ipcMain.handle(IpcChannel.FileService_List, async (_, provider: Provider) => {
+    const service = FileServiceManager.getInstance().getService(provider)
+    return await service.listFiles()
+  })
+
+  ipcMain.handle(IpcChannel.FileService_Delete, async (_, provider: Provider, fileId: string) => {
+    const service = FileServiceManager.getInstance().getService(provider)
+    return await service.deleteFile(fileId)
+  })
+
+  ipcMain.handle(IpcChannel.FileService_Retrieve, async (_, provider: Provider, fileId: string) => {
+    const service = FileServiceManager.getInstance().getService(provider)
+    return await service.retrieveFile(fileId)
+  })
+
   // fs
   ipcMain.handle(IpcChannel.Fs_Read, FileService.readFile)
 
@@ -259,6 +443,7 @@ export function registerIpc(mainWindow: BrowserWindow, app: Electron.App) {
   ipcMain.handle(IpcChannel.KnowledgeBase_Remove, KnowledgeService.remove)
   ipcMain.handle(IpcChannel.KnowledgeBase_Search, KnowledgeService.search)
   ipcMain.handle(IpcChannel.KnowledgeBase_Rerank, KnowledgeService.rerank)
+  ipcMain.handle(IpcChannel.KnowledgeBase_Check_Quota, KnowledgeService.checkQuota)
 
   // window
   ipcMain.handle(IpcChannel.Windows_SetMinimumSize, (_, width: number, height: number) => {
@@ -273,6 +458,15 @@ export function registerIpc(mainWindow: BrowserWindow, app: Electron.App) {
     }
   })
 
+  // VertexAI
+  ipcMain.handle(IpcChannel.VertexAI_GetAuthHeaders, async (_, params) => {
+    return vertexAIService.getAuthHeaders(params)
+  })
+
+  ipcMain.handle(IpcChannel.VertexAI_ClearAuthCache, async (_, projectId: string, clientEmail?: string) => {
+    vertexAIService.clearAuthCache(projectId, clientEmail)
+  })
+
   // mini window
   ipcMain.handle(IpcChannel.MiniWindow_Show, () => windowService.showMiniWindow())
   ipcMain.handle(IpcChannel.MiniWindow_Hide, () => windowService.hideMiniWindow())
@@ -301,6 +495,14 @@ export function registerIpc(mainWindow: BrowserWindow, app: Electron.App) {
   ipcMain.handle(IpcChannel.Mcp_GetInstallInfo, mcpService.getInstallInfo)
   ipcMain.handle(IpcChannel.Mcp_CheckConnectivity, mcpService.checkMcpConnectivity)
 
+  // Register Python execution handler
+  ipcMain.handle(
+    IpcChannel.Python_Execute,
+    async (_, script: string, context?: Record<string, any>, timeout?: number) => {
+      return await pythonService.executeScript(script, context, timeout)
+    }
+  )
+
   ipcMain.handle(IpcChannel.App_IsBinaryExist, (_, name: string) => isBinaryExists(name))
   ipcMain.handle(IpcChannel.App_GetBinaryPath, (_, name: string) => getBinaryPath(name))
   ipcMain.handle(IpcChannel.App_InstallUvBinary, () => runInstallScript('install-uv.js'))
@@ -346,6 +548,12 @@ export function registerIpc(mainWindow: BrowserWindow, app: Electron.App) {
     setOpenLinkExternal(webviewId, isExternal)
   )
 
+  ipcMain.handle(IpcChannel.Webview_SetSpellCheckEnabled, (_, webviewId: number, isEnable: boolean) => {
+    const webview = webContents.fromId(webviewId)
+    if (!webview) return
+    webview.session.setSpellCheckerEnabled(isEnable)
+  })
+
   // store sync
   storeSyncService.registerIpcHandler()
 

src/main/knowledage/embeddings/Embeddings.ts

@@ -5,8 +5,15 @@ import EmbeddingsFactory from './EmbeddingsFactory'
 
 export default class Embeddings {
   private sdk: BaseEmbeddings
-  constructor({ model, apiKey, apiVersion, baseURL, dimensions }: KnowledgeBaseParams) {
-    this.sdk = EmbeddingsFactory.create({ model, apiKey, apiVersion, baseURL, dimensions } as KnowledgeBaseParams)
+  constructor({ model, provider, apiKey, apiVersion, baseURL, dimensions }: KnowledgeBaseParams) {
+    this.sdk = EmbeddingsFactory.create({
+      model,
+      provider,
+      apiKey,
+      apiVersion,
+      baseURL,
+      dimensions
+    } as KnowledgeBaseParams)
   }
   public async init(): Promise<void> {
     return this.sdk.init()

src/main/knowledage/embeddings/EmbeddingsFactory.ts

@@ -0,0 +1,67 @@
+import type { BaseEmbeddings } from '@cherrystudio/embedjs-interfaces'
+import { OllamaEmbeddings } from '@cherrystudio/embedjs-ollama'
+import { OpenAiEmbeddings } from '@cherrystudio/embedjs-openai'
+import { AzureOpenAiEmbeddings } from '@cherrystudio/embedjs-openai/src/azure-openai-embeddings'
+import { getInstanceName } from '@main/utils'
+import { KnowledgeBaseParams } from '@types'
+
+import { SUPPORTED_DIM_MODELS as VOYAGE_SUPPORTED_DIM_MODELS, VoyageEmbeddings } from './VoyageEmbeddings'
+
+export default class EmbeddingsFactory {
+  static create({ model, provider, apiKey, apiVersion, baseURL, dimensions }: KnowledgeBaseParams): BaseEmbeddings {
+    const batchSize = 10
+    if (provider === 'voyageai') {
+      if (VOYAGE_SUPPORTED_DIM_MODELS.includes(model)) {
+        return new VoyageEmbeddings({
+          modelName: model,
+          apiKey,
+          outputDimension: dimensions,
+          batchSize: 8
+        })
+      } else {
+        return new VoyageEmbeddings({
+          modelName: model,
+          apiKey,
+          batchSize: 8
+        })
+      }
+    }
+    if (provider === 'ollama') {
+      if (baseURL.includes('v1/')) {
+        return new OllamaEmbeddings({
+          model: model,
+          baseUrl: baseURL.replace('v1/', ''),
+          requestOptions: {
+            // @ts-ignore expected
+            'encoding-format': 'float'
+          }
+        })
+      }
+      return new OllamaEmbeddings({
+        model: model,
+        baseUrl: baseURL,
+        requestOptions: {
+          // @ts-ignore expected
+          'encoding-format': 'float'
+        }
+      })
+    }
+    if (apiVersion !== undefined) {
+      return new AzureOpenAiEmbeddings({
+        azureOpenAIApiKey: apiKey,
+        azureOpenAIApiVersion: apiVersion,
+        azureOpenAIApiDeploymentName: model,
+        azureOpenAIApiInstanceName: getInstanceName(baseURL),
+        dimensions,
+        batchSize
+      })
+    }
+    return new OpenAiEmbeddings({
+      model,
+      apiKey,
+      dimensions,
+      batchSize,
+      configuration: { baseURL }
+    })
+  }
+}

src/main/knowledage/embeddings/VoyageEmbeddings.ts

@@ -1,16 +1,20 @@
 import { BaseEmbeddings } from '@cherrystudio/embedjs-interfaces'
 import { VoyageEmbeddings as _VoyageEmbeddings } from '@langchain/community/embeddings/voyage'
 
-export default class VoyageEmbeddings extends BaseEmbeddings {
+/**
+ * 支持设置嵌入维度的模型
+ */
+export const SUPPORTED_DIM_MODELS = ['voyage-3-large', 'voyage-3.5', 'voyage-3.5-lite', 'voyage-code-3']
+export class VoyageEmbeddings extends BaseEmbeddings {
   private model: _VoyageEmbeddings
   constructor(private readonly configuration?: ConstructorParameters<typeof _VoyageEmbeddings>[0]) {
     super()
     if (!this.configuration) this.configuration = {}
     if (!this.configuration.modelName) this.configuration.modelName = 'voyage-3'
-
-    if (!this.configuration.outputDimension) {
-      throw new Error('You need to pass in the optional dimensions parameter for this model')
+    if (!SUPPORTED_DIM_MODELS.includes(this.configuration.modelName) && this.configuration.outputDimension) {
+      throw new Error(`VoyageEmbeddings only supports ${SUPPORTED_DIM_MODELS.join(', ')}`)
     }
+
     this.model = new _VoyageEmbeddings(this.configuration)
   }
   override async getDimensions(): Promise<number> {

src/main/knowledage/loader/index.ts

@@ -4,7 +4,7 @@ import { JsonLoader, LocalPathLoader, RAGApplication, TextLoader } from '@cherry
 import type { AddLoaderReturn } from '@cherrystudio/embedjs-interfaces'
 import { WebLoader } from '@cherrystudio/embedjs-loader-web'
 import { LoaderReturn } from '@shared/config/types'
-import { FileType, KnowledgeBaseParams } from '@types'
+import { FileMetadata, KnowledgeBaseParams } from '@types'
 import Logger from 'electron-log'
 
 import { DraftsExportLoader } from './draftsExportLoader'
@@ -16,6 +16,7 @@ const FILE_LOADER_MAP: Record<string, string> = {
   // 内置类型
   '.pdf': 'common',
   '.csv': 'common',
+  '.doc': 'common',
   '.docx': 'common',
   '.pptx': 'common',
   '.xlsx': 'common',
@@ -38,7 +39,7 @@ const FILE_LOADER_MAP: Record<string, string> = {
 
 export async function addOdLoader(
   ragApplication: RAGApplication,
-  file: FileType,
+  file: FileMetadata,
   base: KnowledgeBaseParams,
   forceReload: boolean
 ): Promise<AddLoaderReturn> {
@@ -64,7 +65,7 @@ export async function addOdLoader(
 
 export async function addFileLoader(
   ragApplication: RAGApplication,
-  file: FileType,
+  file: FileMetadata,
   base: KnowledgeBaseParams,
   forceReload: boolean
 ): Promise<LoaderReturn> {

src/main/knowledage/loader/noteLoader.ts

@@ -0,0 +1,44 @@
+import { BaseLoader } from '@cherrystudio/embedjs-interfaces'
+import { cleanString } from '@cherrystudio/embedjs-utils'
+import { RecursiveCharacterTextSplitter } from '@langchain/textsplitters'
+import md5 from 'md5'
+
+export class NoteLoader extends BaseLoader<{ type: 'NoteLoader' }> {
+  private readonly text: string
+  private readonly sourceUrl?: string
+
+  constructor({
+    text,
+    sourceUrl,
+    chunkSize,
+    chunkOverlap
+  }: {
+    text: string
+    sourceUrl?: string
+    chunkSize?: number
+    chunkOverlap?: number
+  }) {
+    super(`NoteLoader_${md5(text + (sourceUrl || ''))}`, { text, sourceUrl }, chunkSize ?? 2000, chunkOverlap ?? 0)
+    this.text = text
+    this.sourceUrl = sourceUrl
+  }
+
+  override async *getUnfilteredChunks() {
+    const chunker = new RecursiveCharacterTextSplitter({
+      chunkSize: this.chunkSize,
+      chunkOverlap: this.chunkOverlap
+    })
+
+    const chunks = await chunker.splitText(cleanString(this.text))
+
+    for (const chunk of chunks) {
+      yield {
+        pageContent: chunk,
+        metadata: {
+          type: 'NoteLoader' as const,
+          source: this.sourceUrl || 'note'
+        }
+      }
+    }
+  }
+}

src/main/knowledage/reranker/BaseReranker.ts

@@ -17,14 +17,17 @@ export default abstract class BaseReranker {
    * Get Rerank Request Url
    */
   protected getRerankUrl() {
-    if (this.base.rerankModelProvider === 'dashscope') {
+    if (this.base.rerankModelProvider === 'bailian') {
       return 'https://dashscope.aliyuncs.com/api/v1/services/rerank/text-rerank/text-rerank'
     }
 
-    let baseURL = this.base?.rerankBaseURL?.endsWith('/')
-      ? this.base.rerankBaseURL.slice(0, -1)
-      : this.base.rerankBaseURL
-    // 必须携带/v1，否则会404
+    let baseURL = this.base.rerankBaseURL
+
+    if (baseURL && baseURL.endsWith('/')) {
+      // `/` 结尾强制使用rerankBaseURL
+      return `${baseURL}rerank`
+    }
+
     if (baseURL && !baseURL.endsWith('/v1')) {
       baseURL = `${baseURL}/v1`
     }
@@ -47,7 +50,7 @@ export default abstract class BaseReranker {
         documents,
         top_k: topN
       }
-    } else if (provider === 'dashscope') {
+    } else if (provider === 'bailian') {
       return {
         model: this.base.rerankModel,
         input: {
@@ -58,6 +61,12 @@ export default abstract class BaseReranker {
           top_n: topN
         }
       }
+    } else if (provider?.includes('tei')) {
+      return {
+        query,
+        texts: documents,
+        return_text: true
+      }
     } else {
       return {
         model: this.base.rerankModel,
@@ -73,10 +82,17 @@ export default abstract class BaseReranker {
    */
   protected extractRerankResult(data: any) {
     const provider = this.base.rerankModelProvider
-    if (provider === 'dashscope') {
+    if (provider === 'bailian') {
       return data.output.results
     } else if (provider === 'voyageai') {
       return data.data
+    } else if (provider?.includes('tei')) {
+      return data.map((item: any) => {
+        return {
+          index: item.index,
+          relevance_score: item.score
+        }
+      })
     } else {
       return data.results
     }

src/main/mcpServers/factory.ts

@@ -6,6 +6,7 @@ import DifyKnowledgeServer from './dify-knowledge'
 import FetchServer from './fetch'
 import FileSystemServer from './filesystem'
 import MemoryServer from './memory'
+import PythonServer from './python'
 import ThinkingServer from './sequentialthinking'
 
 export function createInMemoryMCPServer(name: string, args: string[] = [], envs: Record<string, string> = {}): Server {
@@ -31,6 +32,9 @@ export function createInMemoryMCPServer(name: string, args: string[] = [], envs:
       const difyKey = envs.DIFY_KEY
       return new DifyKnowledgeServer(difyKey, args).server
     }
+    case '@cherry/python': {
+      return new PythonServer().server
+    }
     default:
       throw new Error(`Unknown in-memory MCP server: ${name}`)
   }

src/main/mcpServers/python.ts

@@ -0,0 +1,113 @@
+import { pythonService } from '@main/services/PythonService'
+import { Server } from '@modelcontextprotocol/sdk/server/index.js'
+import { CallToolRequestSchema, ErrorCode, ListToolsRequestSchema, McpError } from '@modelcontextprotocol/sdk/types.js'
+import Logger from 'electron-log'
+
+/**
+ * Python MCP Server for executing Python code using Pyodide
+ */
+class PythonServer {
+  public server: Server
+
+  constructor() {
+    this.server = new Server(
+      {
+        name: 'python-server',
+        version: '1.0.0'
+      },
+      {
+        capabilities: {
+          tools: {}
+        }
+      }
+    )
+
+    this.setupRequestHandlers()
+  }
+
+  private setupRequestHandlers() {
+    // List available tools
+    this.server.setRequestHandler(ListToolsRequestSchema, async () => {
+      return {
+        tools: [
+          {
+            name: 'python_execute',
+            description: `Execute Python code using Pyodide in a sandboxed environment. Supports most Python standard library and scientific packages.
+The code will be executed with Python 3.12.
+Dependencies may be defined via PEP 723 script metadata, e.g. to install "pydantic", the script should start
+with a comment of the form:
+# /// script
+# dependencies = ['pydantic']
+# ///
+print('python code here')`,
+            inputSchema: {
+              type: 'object',
+              properties: {
+                code: {
+                  type: 'string',
+                  description: 'The Python code to execute'
+                },
+                context: {
+                  type: 'object',
+                  description: 'Optional context variables to pass to the Python execution environment',
+                  additionalProperties: true
+                },
+                timeout: {
+                  type: 'number',
+                  description: 'Timeout in milliseconds (default: 60000)',
+                  default: 60000
+                }
+              },
+              required: ['code']
+            }
+          }
+        ]
+      }
+    })
+
+    // Handle tool calls
+    this.server.setRequestHandler(CallToolRequestSchema, async (request) => {
+      const { name, arguments: args } = request.params
+
+      if (name !== 'python_execute') {
+        throw new McpError(ErrorCode.MethodNotFound, `Tool ${name} not found`)
+      }
+
+      try {
+        const {
+          code,
+          context = {},
+          timeout = 60000
+        } = args as {
+          code: string
+          context?: Record<string, any>
+          timeout?: number
+        }
+
+        if (!code || typeof code !== 'string') {
+          throw new McpError(ErrorCode.InvalidParams, 'Code parameter is required and must be a string')
+        }
+
+        Logger.info('Executing Python code via Pyodide')
+
+        const result = await pythonService.executeScript(code, context, timeout)
+
+        return {
+          content: [
+            {
+              type: 'text',
+              text: result
+            }
+          ]
+        }
+      } catch (error) {
+        const errorMessage = error instanceof Error ? error.message : String(error)
+        Logger.error('Python execution error:', errorMessage)
+
+        throw new McpError(ErrorCode.InternalError, `Python execution failed: ${errorMessage}`)
+      }
+    })
+  }
+}
+
+export default PythonServer

src/main/mcpServers/sequentialthinking.ts

@@ -106,6 +106,7 @@ class SequentialThinkingServer {
             type: 'text',
             text: JSON.stringify(
               {
+                thought: validatedInput.thought,
                 thoughtNumber: validatedInput.thoughtNumber,
                 totalThoughts: validatedInput.totalThoughts,
                 nextThoughtNeeded: validatedInput.nextThoughtNeeded,

src/main/ocr/BaseOcrProvider.ts

@@ -0,0 +1,122 @@
+import fs from 'node:fs'
+import path from 'node:path'
+
+import { windowService } from '@main/services/WindowService'
+import { getFileExt } from '@main/utils/file'
+import { FileMetadata, OcrProvider } from '@types'
+import { app } from 'electron'
+import { TypedArray } from 'pdfjs-dist/types/src/display/api'
+
+export default abstract class BaseOcrProvider {
+  protected provider: OcrProvider
+  public storageDir = path.join(app.getPath('userData'), 'Data', 'Files')
+
+  constructor(provider: OcrProvider) {
+    if (!provider) {
+      throw new Error('OCR provider is not set')
+    }
+    this.provider = provider
+  }
+  abstract parseFile(sourceId: string, file: FileMetadata): Promise<{ processedFile: FileMetadata; quota?: number }>
+
+  /**
+   * 检查文件是否已经被预处理过
+   * 统一检测方法：如果 Data/Files/{file.id} 是目录，说明已被预处理
+   * @param file 文件信息
+   * @returns 如果已处理返回处理后的文件信息，否则返回null
+   */
+  public async checkIfAlreadyProcessed(file: FileMetadata): Promise<FileMetadata | null> {
+    try {
+      // 检查 Data/Files/{file.id} 是否是目录
+      const preprocessDirPath = path.join(this.storageDir, file.id)
+
+      if (fs.existsSync(preprocessDirPath)) {
+        const stats = await fs.promises.stat(preprocessDirPath)
+
+        // 如果是目录，说明已经被预处理过
+        if (stats.isDirectory()) {
+          // 查找目录中的处理结果文件
+          const files = await fs.promises.readdir(preprocessDirPath)
+
+          // 查找主要的处理结果文件（.md 或 .txt）
+          const processedFile = files.find((fileName) => fileName.endsWith('.md') || fileName.endsWith('.txt'))
+
+          if (processedFile) {
+            const processedFilePath = path.join(preprocessDirPath, processedFile)
+            const processedStats = await fs.promises.stat(processedFilePath)
+            const ext = getFileExt(processedFile)
+
+            return {
+              ...file,
+              name: file.name.replace(file.ext, ext),
+              path: processedFilePath,
+              ext: ext,
+              size: processedStats.size,
+              created_at: processedStats.birthtime.toISOString()
+            }
+          }
+        }
+      }
+
+      return null
+    } catch (error) {
+      // 如果检查过程中出现错误，返回null表示未处理
+      return null
+    }
+  }
+
+  /**
+   * 辅助方法：延迟执行
+   */
+  public delay = (ms: number): Promise<void> => {
+    return new Promise((resolve) => setTimeout(resolve, ms))
+  }
+
+  public async readPdf(
+    source: string | URL | TypedArray,
+    passwordCallback?: (fn: (password: string) => void, reason: string) => string
+  ) {
+    const { getDocument } = await import('pdfjs-dist/legacy/build/pdf.mjs')
+    const documentLoadingTask = getDocument(source)
+    if (passwordCallback) {
+      documentLoadingTask.onPassword = passwordCallback
+    }
+
+    const document = await documentLoadingTask.promise
+    return document
+  }
+
+  public async sendOcrProgress(sourceId: string, progress: number): Promise<void> {
+    const mainWindow = windowService.getMainWindow()
+    mainWindow?.webContents.send('file-ocr-progress', {
+      itemId: sourceId,
+      progress: progress
+    })
+  }
+
+  /**
+   * 将文件移动到附件目录
+   * @param fileId 文件id
+   * @param filePaths 需要移动的文件路径数组
+   * @returns 移动后的文件路径数组
+   */
+  public moveToAttachmentsDir(fileId: string, filePaths: string[]): string[] {
+    const attachmentsPath = path.join(this.storageDir, fileId)
+    if (!fs.existsSync(attachmentsPath)) {
+      fs.mkdirSync(attachmentsPath, { recursive: true })
+    }
+
+    const movedPaths: string[] = []
+
+    for (const filePath of filePaths) {
+      if (fs.existsSync(filePath)) {
+        const fileName = path.basename(filePath)
+        const destPath = path.join(attachmentsPath, fileName)
+        fs.copyFileSync(filePath, destPath)
+        fs.unlinkSync(filePath) // 删除原文件，实现"移动"
+        movedPaths.push(destPath)
+      }
+    }
+    return movedPaths
+  }
+}

src/main/ocr/DefaultOcrProvider.ts

@@ -0,0 +1,12 @@
+import { FileMetadata, OcrProvider } from '@types'
+
+import BaseOcrProvider from './BaseOcrProvider'
+
+export default class DefaultOcrProvider extends BaseOcrProvider {
+  constructor(provider: OcrProvider) {
+    super(provider)
+  }
+  public parseFile(): Promise<{ processedFile: FileMetadata }> {
+    throw new Error('Method not implemented.')
+  }
+}

src/main/ocr/MacSysOcrProvider.ts

@@ -0,0 +1,128 @@
+import { isMac } from '@main/constant'
+import { FileMetadata, OcrProvider } from '@types'
+import Logger from 'electron-log'
+import * as fs from 'fs'
+import * as path from 'path'
+import { TextItem } from 'pdfjs-dist/types/src/display/api'
+
+import BaseOcrProvider from './BaseOcrProvider'
+
+export default class MacSysOcrProvider extends BaseOcrProvider {
+  private readonly MIN_TEXT_LENGTH = 1000
+  private MacOCR: any
+
+  private async initMacOCR() {
+    if (!isMac) {
+      throw new Error('MacSysOcrProvider is only available on macOS')
+    }
+    if (!this.MacOCR) {
+      try {
+        // @ts-ignore This module is optional and only installed/available on macOS. Runtime checks prevent execution on other platforms.
+        const module = await import('@cherrystudio/mac-system-ocr')
+        this.MacOCR = module.default
+      } catch (error) {
+        Logger.error('[OCR] Failed to load mac-system-ocr:', error)
+        throw error
+      }
+    }
+    return this.MacOCR
+  }
+
+  private getRecognitionLevel(level?: number) {
+    return level === 0 ? this.MacOCR.RECOGNITION_LEVEL_FAST : this.MacOCR.RECOGNITION_LEVEL_ACCURATE
+  }
+
+  constructor(provider: OcrProvider) {
+    super(provider)
+  }
+
+  private async processPages(
+    results: any,
+    totalPages: number,
+    sourceId: string,
+    writeStream: fs.WriteStream
+  ): Promise<void> {
+    await this.initMacOCR()
+    // TODO: 下个版本后面使用批处理，以及p-queue来优化
+    for (let i = 0; i < totalPages; i++) {
+      // Convert pages to buffers
+      const pageNum = i + 1
+      const pageBuffer = await results.getPage(pageNum)
+
+      // Process batch
+      const ocrResult = await this.MacOCR.recognizeFromBuffer(pageBuffer, {
+        ocrOptions: {
+          recognitionLevel: this.getRecognitionLevel(this.provider.options?.recognitionLevel),
+          minConfidence: this.provider.options?.minConfidence || 0.5
+        }
+      })
+
+      // Write results in order
+      writeStream.write(ocrResult.text + '\n')
+
+      // Update progress
+      await this.sendOcrProgress(sourceId, (pageNum / totalPages) * 100)
+    }
+  }
+
+  public async isScanPdf(buffer: Buffer): Promise<boolean> {
+    const doc = await this.readPdf(new Uint8Array(buffer))
+    const pageLength = doc.numPages
+    let counts = 0
+    const pagesToCheck = Math.min(pageLength, 10)
+    for (let i = 0; i < pagesToCheck; i++) {
+      const page = await doc.getPage(i + 1)
+      const pageData = await page.getTextContent()
+      const pageText = pageData.items.map((item) => (item as TextItem).str).join('')
+      counts += pageText.length
+      if (counts >= this.MIN_TEXT_LENGTH) {
+        return false
+      }
+    }
+    return true
+  }
+
+  public async parseFile(sourceId: string, file: FileMetadata): Promise<{ processedFile: FileMetadata }> {
+    Logger.info(`[OCR] Starting OCR process for file: ${file.name}`)
+    if (file.ext === '.pdf') {
+      try {
+        const { pdf } = await import('@cherrystudio/pdf-to-img-napi')
+        const pdfBuffer = await fs.promises.readFile(file.path)
+        const results = await pdf(pdfBuffer, {
+          scale: 2
+        })
+        const totalPages = results.length
+
+        const baseDir = path.dirname(file.path)
+        const baseName = path.basename(file.path, path.extname(file.path))
+        const txtFileName = `${baseName}.txt`
+        const txtFilePath = path.join(baseDir, txtFileName)
+
+        const writeStream = fs.createWriteStream(txtFilePath)
+        await this.processPages(results, totalPages, sourceId, writeStream)
+
+        await new Promise<void>((resolve, reject) => {
+          writeStream.end(() => {
+            Logger.info(`[OCR] OCR process completed successfully for ${file.origin_name}`)
+            resolve()
+          })
+          writeStream.on('error', reject)
+        })
+        const movedPaths = this.moveToAttachmentsDir(file.id, [txtFilePath])
+        return {
+          processedFile: {
+            ...file,
+            name: txtFileName,
+            path: movedPaths[0],
+            ext: '.txt',
+            size: fs.statSync(movedPaths[0]).size
+          }
+        }
+      } catch (error) {
+        Logger.error('[OCR] Error during OCR process:', error)
+        throw error
+      }
+    }
+    return { processedFile: file }
+  }
+}

src/main/ocr/OcrProvider.ts

@@ -0,0 +1,26 @@
+import { FileMetadata, OcrProvider as Provider } from '@types'
+
+import BaseOcrProvider from './BaseOcrProvider'
+import OcrProviderFactory from './OcrProviderFactory'
+
+export default class OcrProvider {
+  private sdk: BaseOcrProvider
+  constructor(provider: Provider) {
+    this.sdk = OcrProviderFactory.create(provider)
+  }
+  public async parseFile(
+    sourceId: string,
+    file: FileMetadata
+  ): Promise<{ processedFile: FileMetadata; quota?: number }> {
+    return this.sdk.parseFile(sourceId, file)
+  }
+
+  /**
+   * 检查文件是否已经被预处理过
+   * @param file 文件信息
+   * @returns 如果已处理返回处理后的文件信息，否则返回null
+   */
+  public async checkIfAlreadyProcessed(file: FileMetadata): Promise<FileMetadata | null> {
+    return this.sdk.checkIfAlreadyProcessed(file)
+  }
+}

src/main/ocr/OcrProviderFactory.ts

@@ -0,0 +1,20 @@
+import { isMac } from '@main/constant'
+import { OcrProvider } from '@types'
+import Logger from 'electron-log'
+
+import BaseOcrProvider from './BaseOcrProvider'
+import DefaultOcrProvider from './DefaultOcrProvider'
+import MacSysOcrProvider from './MacSysOcrProvider'
+export default class OcrProviderFactory {
+  static create(provider: OcrProvider): BaseOcrProvider {
+    switch (provider.id) {
+      case 'system':
+        if (!isMac) {
+          Logger.warn('[OCR] System OCR provider is only available on macOS')
+        }
+        return new MacSysOcrProvider(provider)
+      default:
+        return new DefaultOcrProvider(provider)
+    }
+  }
+}

src/main/preprocess/BasePreprocessProvider.ts

@@ -0,0 +1,126 @@
+import fs from 'node:fs'
+import path from 'node:path'
+
+import { windowService } from '@main/services/WindowService'
+import { getFileExt } from '@main/utils/file'
+import { FileMetadata, PreprocessProvider } from '@types'
+import { app } from 'electron'
+import { TypedArray } from 'pdfjs-dist/types/src/display/api'
+
+export default abstract class BasePreprocessProvider {
+  protected provider: PreprocessProvider
+  protected userId?: string
+  public storageDir = path.join(app.getPath('userData'), 'Data', 'Files')
+
+  constructor(provider: PreprocessProvider, userId?: string) {
+    if (!provider) {
+      throw new Error('Preprocess provider is not set')
+    }
+    this.provider = provider
+    this.userId = userId
+  }
+  abstract parseFile(sourceId: string, file: FileMetadata): Promise<{ processedFile: FileMetadata; quota?: number }>
+
+  abstract checkQuota(): Promise<number>
+
+  /**
+   * 检查文件是否已经被预处理过
+   * 统一检测方法：如果 Data/Files/{file.id} 是目录，说明已被预处理
+   * @param file 文件信息
+   * @returns 如果已处理返回处理后的文件信息，否则返回null
+   */
+  public async checkIfAlreadyProcessed(file: FileMetadata): Promise<FileMetadata | null> {
+    try {
+      // 检查 Data/Files/{file.id} 是否是目录
+      const preprocessDirPath = path.join(this.storageDir, file.id)
+
+      if (fs.existsSync(preprocessDirPath)) {
+        const stats = await fs.promises.stat(preprocessDirPath)
+
+        // 如果是目录，说明已经被预处理过
+        if (stats.isDirectory()) {
+          // 查找目录中的处理结果文件
+          const files = await fs.promises.readdir(preprocessDirPath)
+
+          // 查找主要的处理结果文件（.md 或 .txt）
+          const processedFile = files.find((fileName) => fileName.endsWith('.md') || fileName.endsWith('.txt'))
+
+          if (processedFile) {
+            const processedFilePath = path.join(preprocessDirPath, processedFile)
+            const processedStats = await fs.promises.stat(processedFilePath)
+            const ext = getFileExt(processedFile)
+
+            return {
+              ...file,
+              name: file.name.replace(file.ext, ext),
+              path: processedFilePath,
+              ext: ext,
+              size: processedStats.size,
+              created_at: processedStats.birthtime.toISOString()
+            }
+          }
+        }
+      }
+
+      return null
+    } catch (error) {
+      // 如果检查过程中出现错误，返回null表示未处理
+      return null
+    }
+  }
+
+  /**
+   * 辅助方法：延迟执行
+   */
+  public delay = (ms: number): Promise<void> => {
+    return new Promise((resolve) => setTimeout(resolve, ms))
+  }
+
+  public async readPdf(
+    source: string | URL | TypedArray,
+    passwordCallback?: (fn: (password: string) => void, reason: string) => string
+  ) {
+    const { getDocument } = await import('pdfjs-dist/legacy/build/pdf.mjs')
+    const documentLoadingTask = getDocument(source)
+    if (passwordCallback) {
+      documentLoadingTask.onPassword = passwordCallback
+    }
+
+    const document = await documentLoadingTask.promise
+    return document
+  }
+
+  public async sendPreprocessProgress(sourceId: string, progress: number): Promise<void> {
+    const mainWindow = windowService.getMainWindow()
+    mainWindow?.webContents.send('file-preprocess-progress', {
+      itemId: sourceId,
+      progress: progress
+    })
+  }
+
+  /**
+   * 将文件移动到附件目录
+   * @param fileId 文件id
+   * @param filePaths 需要移动的文件路径数组
+   * @returns 移动后的文件路径数组
+   */
+  public moveToAttachmentsDir(fileId: string, filePaths: string[]): string[] {
+    const attachmentsPath = path.join(this.storageDir, fileId)
+    if (!fs.existsSync(attachmentsPath)) {
+      fs.mkdirSync(attachmentsPath, { recursive: true })
+    }
+
+    const movedPaths: string[] = []
+
+    for (const filePath of filePaths) {
+      if (fs.existsSync(filePath)) {
+        const fileName = path.basename(filePath)
+        const destPath = path.join(attachmentsPath, fileName)
+        fs.copyFileSync(filePath, destPath)
+        fs.unlinkSync(filePath) // 删除原文件，实现"移动"
+        movedPaths.push(destPath)
+      }
+    }
+    return movedPaths
+  }
+}

src/main/preprocess/DefaultPreprocessProvider.ts

@@ -0,0 +1,16 @@
+import { FileMetadata, PreprocessProvider } from '@types'
+
+import BasePreprocessProvider from './BasePreprocessProvider'
+
+export default class DefaultPreprocessProvider extends BasePreprocessProvider {
+  constructor(provider: PreprocessProvider) {
+    super(provider)
+  }
+  public parseFile(): Promise<{ processedFile: FileMetadata }> {
+    throw new Error('Method not implemented.')
+  }
+
+  public checkQuota(): Promise<number> {
+    throw new Error('Method not implemented.')
+  }
+}

src/main/preprocess/Doc2xPreprocessProvider.ts

@@ -0,0 +1,329 @@
+import fs from 'node:fs'
+import path from 'node:path'
+
+import { FileMetadata, PreprocessProvider } from '@types'
+import AdmZip from 'adm-zip'
+import axios, { AxiosRequestConfig } from 'axios'
+import Logger from 'electron-log'
+
+import BasePreprocessProvider from './BasePreprocessProvider'
+
+type ApiResponse<T> = {
+  code: string
+  data: T
+  message?: string
+}
+
+type PreuploadResponse = {
+  uid: string
+  url: string
+}
+
+type StatusResponse = {
+  status: string
+  progress: number
+}
+
+type ParsedFileResponse = {
+  status: string
+  url: string
+}
+
+export default class Doc2xPreprocessProvider extends BasePreprocessProvider {
+  constructor(provider: PreprocessProvider) {
+    super(provider)
+  }
+
+  private async validateFile(filePath: string): Promise<void> {
+    const pdfBuffer = await fs.promises.readFile(filePath)
+
+    const doc = await this.readPdf(new Uint8Array(pdfBuffer))
+
+    // 文件页数小于1000页
+    if (doc.numPages >= 1000) {
+      throw new Error(`PDF page count (${doc.numPages}) exceeds the limit of 1000 pages`)
+    }
+    // 文件大小小于300MB
+    if (pdfBuffer.length >= 300 * 1024 * 1024) {
+      const fileSizeMB = Math.round(pdfBuffer.length / (1024 * 1024))
+      throw new Error(`PDF file size (${fileSizeMB}MB) exceeds the limit of 300MB`)
+    }
+  }
+
+  public async parseFile(sourceId: string, file: FileMetadata): Promise<{ processedFile: FileMetadata }> {
+    try {
+      Logger.info(`Preprocess processing started: ${file.path}`)
+
+      // 步骤1: 准备上传
+      const { uid, url } = await this.preupload()
+      Logger.info(`Preprocess preupload completed: uid=${uid}`)
+
+      await this.validateFile(file.path)
+
+      // 步骤2: 上传文件
+      await this.putFile(file.path, url)
+
+      // 步骤3: 等待处理完成
+      await this.waitForProcessing(sourceId, uid)
+      Logger.info(`Preprocess parsing completed successfully for: ${file.path}`)
+
+      // 步骤4: 导出文件
+      const { path: outputPath } = await this.exportFile(file, uid)
+
+      // 步骤5: 创建处理后的文件信息
+      return {
+        processedFile: this.createProcessedFileInfo(file, outputPath)
+      }
+    } catch (error) {
+      Logger.error(
+        `Preprocess processing failed for ${file.path}: ${error instanceof Error ? error.message : String(error)}`
+      )
+      throw error
+    }
+  }
+
+  private createProcessedFileInfo(file: FileMetadata, outputPath: string): FileMetadata {
+    const outputFilePath = `${outputPath}/${file.name.split('.').slice(0, -1).join('.')}.md`
+    return {
+      ...file,
+      name: file.name.replace('.pdf', '.md'),
+      path: outputFilePath,
+      ext: '.md',
+      size: fs.statSync(outputFilePath).size
+    }
+  }
+
+  /**
+   * 导出文件
+   * @param file 文件信息
+   * @param uid 预上传响应的uid
+   * @returns 导出文件的路径
+   */
+  public async exportFile(file: FileMetadata, uid: string): Promise<{ path: string }> {
+    Logger.info(`Exporting file: ${file.path}`)
+
+    // 步骤1: 转换文件
+    await this.convertFile(uid, file.path)
+    Logger.info(`File conversion completed for: ${file.path}`)
+
+    // 步骤2: 等待导出并获取URL
+    const exportUrl = await this.waitForExport(uid)
+
+    // 步骤3: 下载并解压文件
+    return this.downloadFile(exportUrl, file)
+  }
+
+  /**
+   * 等待处理完成
+   * @param sourceId 源文件ID
+   * @param uid 预上传响应的uid
+   */
+  private async waitForProcessing(sourceId: string, uid: string): Promise<void> {
+    while (true) {
+      await this.delay(1000)
+      const { status, progress } = await this.getStatus(uid)
+      await this.sendPreprocessProgress(sourceId, progress)
+      Logger.info(`Preprocess processing status: ${status}, progress: ${progress}%`)
+
+      if (status === 'success') {
+        return
+      } else if (status === 'failed') {
+        throw new Error('Preprocess processing failed')
+      }
+    }
+  }
+
+  /**
+   * 等待导出完成
+   * @param uid 预上传响应的uid
+   * @returns 导出文件的url
+   */
+  private async waitForExport(uid: string): Promise<string> {
+    while (true) {
+      await this.delay(1000)
+      const { status, url } = await this.getParsedFile(uid)
+      Logger.info(`Export status: ${status}`)
+
+      if (status === 'success' && url) {
+        return url
+      } else if (status === 'failed') {
+        throw new Error('Export failed')
+      }
+    }
+  }
+
+  /**
+   * 预上传文件
+   * @returns 预上传响应的url和uid
+   */
+  private async preupload(): Promise<PreuploadResponse> {
+    const config = this.createAuthConfig()
+    const endpoint = `${this.provider.apiHost}/api/v2/parse/preupload`
+
+    try {
+      const { data } = await axios.post<ApiResponse<PreuploadResponse>>(endpoint, null, config)
+
+      if (data.code === 'success' && data.data) {
+        return data.data
+      } else {
+        throw new Error(`API returned error: ${data.message || JSON.stringify(data)}`)
+      }
+    } catch (error) {
+      Logger.error(`Failed to get preupload URL: ${error instanceof Error ? error.message : String(error)}`)
+      throw new Error('Failed to get preupload URL')
+    }
+  }
+
+  /**
+   * 上传文件
+   * @param filePath 文件路径
+   * @param url 预上传响应的url
+   */
+  private async putFile(filePath: string, url: string): Promise<void> {
+    try {
+      const fileStream = fs.createReadStream(filePath)
+      const response = await axios.put(url, fileStream)
+
+      if (response.status !== 200) {
+        throw new Error(`HTTP status ${response.status}: ${response.statusText}`)
+      }
+    } catch (error) {
+      Logger.error(`Failed to upload file ${filePath}: ${error instanceof Error ? error.message : String(error)}`)
+      throw new Error('Failed to upload file')
+    }
+  }
+
+  private async getStatus(uid: string): Promise<StatusResponse> {
+    const config = this.createAuthConfig()
+    const endpoint = `${this.provider.apiHost}/api/v2/parse/status?uid=${uid}`
+
+    try {
+      const response = await axios.get<ApiResponse<StatusResponse>>(endpoint, config)
+
+      if (response.data.code === 'success' && response.data.data) {
+        return response.data.data
+      } else {
+        throw new Error(`API returned error: ${response.data.message || JSON.stringify(response.data)}`)
+      }
+    } catch (error) {
+      Logger.error(`Failed to get status for uid ${uid}: ${error instanceof Error ? error.message : String(error)}`)
+      throw new Error('Failed to get processing status')
+    }
+  }
+
+  /**
+   * Preprocess文件
+   * @param uid 预上传响应的uid
+   * @param filePath 文件路径
+   */
+  private async convertFile(uid: string, filePath: string): Promise<void> {
+    const fileName = path.basename(filePath).split('.')[0]
+    const config = {
+      ...this.createAuthConfig(),
+      headers: {
+        ...this.createAuthConfig().headers,
+        'Content-Type': 'application/json'
+      }
+    }
+
+    const payload = {
+      uid,
+      to: 'md',
+      formula_mode: 'normal',
+      filename: fileName
+    }
+
+    const endpoint = `${this.provider.apiHost}/api/v2/convert/parse`
+
+    try {
+      const response = await axios.post<ApiResponse<any>>(endpoint, payload, config)
+
+      if (response.data.code !== 'success') {
+        throw new Error(`API returned error: ${response.data.message || JSON.stringify(response.data)}`)
+      }
+    } catch (error) {
+      Logger.error(`Failed to convert file ${filePath}: ${error instanceof Error ? error.message : String(error)}`)
+      throw new Error('Failed to convert file')
+    }
+  }
+
+  /**
+   * 获取解析后的文件信息
+   * @param uid 预上传响应的uid
+   * @returns 解析后的文件信息
+   */
+  private async getParsedFile(uid: string): Promise<ParsedFileResponse> {
+    const config = this.createAuthConfig()
+    const endpoint = `${this.provider.apiHost}/api/v2/convert/parse/result?uid=${uid}`
+
+    try {
+      const response = await axios.get<ApiResponse<ParsedFileResponse>>(endpoint, config)
+
+      if (response.status === 200 && response.data.data) {
+        return response.data.data
+      } else {
+        throw new Error(`HTTP status ${response.status}: ${response.statusText}`)
+      }
+    } catch (error) {
+      Logger.error(
+        `Failed to get parsed file for uid ${uid}: ${error instanceof Error ? error.message : String(error)}`
+      )
+      throw new Error('Failed to get parsed file information')
+    }
+  }
+
+  /**
+   * 下载文件
+   * @param url 导出文件的url
+   * @param file 文件信息
+   * @returns 下载文件的路径
+   */
+  private async downloadFile(url: string, file: FileMetadata): Promise<{ path: string }> {
+    const dirPath = this.storageDir
+    // 使用统一的存储路径：Data/Files/{file.id}/
+    const extractPath = path.join(dirPath, file.id)
+    const zipPath = path.join(dirPath, `${file.id}.zip`)
+
+    // 确保目录存在
+    fs.mkdirSync(dirPath, { recursive: true })
+    fs.mkdirSync(extractPath, { recursive: true })
+
+    Logger.info(`Downloading to export path: ${zipPath}`)
+
+    try {
+      // 下载文件
+      const response = await axios.get(url, { responseType: 'arraybuffer' })
+      fs.writeFileSync(zipPath, response.data)
+
+      // 确保提取目录存在
+      if (!fs.existsSync(extractPath)) {
+        fs.mkdirSync(extractPath, { recursive: true })
+      }
+
+      // 解压文件
+      const zip = new AdmZip(zipPath)
+      zip.extractAllTo(extractPath, true)
+      Logger.info(`Extracted files to: ${extractPath}`)
+
+      // 删除临时ZIP文件
+      fs.unlinkSync(zipPath)
+
+      return { path: extractPath }
+    } catch (error) {
+      Logger.error(`Failed to download and extract file: ${error instanceof Error ? error.message : String(error)}`)
+      throw new Error('Failed to download and extract file')
+    }
+  }
+
+  private createAuthConfig(): AxiosRequestConfig {
+    return {
+      headers: {
+        Authorization: `Bearer ${this.provider.apiKey}`
+      }
+    }
+  }
+
+  public checkQuota(): Promise<number> {
+    throw new Error('Method not implemented.')
+  }
+}

src/main/preprocess/MineruPreprocessProvider.ts

@@ -0,0 +1,399 @@
+import fs from 'node:fs'
+import path from 'node:path'
+
+import { FileMetadata, PreprocessProvider } from '@types'
+import AdmZip from 'adm-zip'
+import axios from 'axios'
+import Logger from 'electron-log'
+
+import BasePreprocessProvider from './BasePreprocessProvider'
+
+type ApiResponse<T> = {
+  code: number
+  data: T
+  msg?: string
+  trace_id?: string
+}
+
+type BatchUploadResponse = {
+  batch_id: string
+  file_urls: string[]
+}
+
+type ExtractProgress = {
+  extracted_pages: number
+  total_pages: number
+  start_time: string
+}
+
+type ExtractFileResult = {
+  file_name: string
+  state: 'done' | 'waiting-file' | 'pending' | 'running' | 'converting' | 'failed'
+  err_msg: string
+  full_zip_url?: string
+  extract_progress?: ExtractProgress
+}
+
+type ExtractResultResponse = {
+  batch_id: string
+  extract_result: ExtractFileResult[]
+}
+
+type QuotaResponse = {
+  code: number
+  data: {
+    user_left_quota: number
+    total_left_quota: number
+  }
+  msg?: string
+  trace_id?: string
+}
+
+export default class MineruPreprocessProvider extends BasePreprocessProvider {
+  constructor(provider: PreprocessProvider, userId?: string) {
+    super(provider, userId)
+    // todo：免费期结束后删除
+    this.provider.apiKey = this.provider.apiKey || import.meta.env.MAIN_VITE_MINERU_API_KEY
+  }
+
+  public async parseFile(
+    sourceId: string,
+    file: FileMetadata
+  ): Promise<{ processedFile: FileMetadata; quota: number }> {
+    try {
+      Logger.info(`MinerU preprocess processing started: ${file.path}`)
+      await this.validateFile(file.path)
+
+      // 1. 获取上传URL并上传文件
+      const batchId = await this.uploadFile(file)
+      Logger.info(`MinerU file upload completed: batch_id=${batchId}`)
+
+      // 2. 等待处理完成并获取结果
+      const extractResult = await this.waitForCompletion(sourceId, batchId, file.origin_name)
+      Logger.info(`MinerU processing completed for batch: ${batchId}`)
+
+      // 3. 下载并解压文件
+      const { path: outputPath } = await this.downloadAndExtractFile(extractResult.full_zip_url!, file)
+
+      // 4. check quota
+      const quota = await this.checkQuota()
+
+      // 5. 创建处理后的文件信息
+      return {
+        processedFile: this.createProcessedFileInfo(file, outputPath),
+        quota
+      }
+    } catch (error: any) {
+      Logger.error(`MinerU preprocess processing failed for ${file.path}: ${error.message}`)
+      throw new Error(error.message)
+    }
+  }
+
+  public async checkQuota() {
+    try {
+      const quota = await fetch(`${this.provider.apiHost}/api/v4/quota`, {
+        method: 'GET',
+        headers: {
+          'Content-Type': 'application/json',
+          Authorization: `Bearer ${this.provider.apiKey}`,
+          token: this.userId ?? ''
+        }
+      })
+      if (!quota.ok) {
+        throw new Error(`HTTP ${quota.status}: ${quota.statusText}`)
+      }
+      const response: QuotaResponse = await quota.json()
+      return response.data.user_left_quota
+    } catch (error) {
+      console.error('Error checking quota:', error)
+      throw error
+    }
+  }
+
+  private async validateFile(filePath: string): Promise<void> {
+    const quota = await this.checkQuota()
+    const pdfBuffer = await fs.promises.readFile(filePath)
+
+    const doc = await this.readPdf(new Uint8Array(pdfBuffer))
+
+    // 文件页数小于600页
+    if (doc.numPages >= 600) {
+      throw new Error(`PDF page count (${doc.numPages}) exceeds the limit of 600 pages`)
+    }
+    // 文件大小小于200MB
+    if (pdfBuffer.length >= 200 * 1024 * 1024) {
+      const fileSizeMB = Math.round(pdfBuffer.length / (1024 * 1024))
+      throw new Error(`PDF file size (${fileSizeMB}MB) exceeds the limit of 200MB`)
+    }
+    // 检查配额
+    if (quota <= 0 || quota - doc.numPages <= 0) {
+      throw new Error('MinerU解析配额不足，请申请企业账户或自行部署，剩余额度：' + quota)
+    }
+  }
+
+  private createProcessedFileInfo(file: FileMetadata, outputPath: string): FileMetadata {
+    // 查找解压后的主要文件
+    let finalPath = ''
+    let finalName = file.origin_name.replace('.pdf', '.md')
+
+    try {
+      const files = fs.readdirSync(outputPath)
+
+      const mdFile = files.find((f) => f.endsWith('.md'))
+      if (mdFile) {
+        const originalMdPath = path.join(outputPath, mdFile)
+        const newMdPath = path.join(outputPath, finalName)
+
+        // 重命名文件为原始文件名
+        try {
+          fs.renameSync(originalMdPath, newMdPath)
+          finalPath = newMdPath
+          Logger.info(`Renamed markdown file from ${mdFile} to ${finalName}`)
+        } catch (renameError) {
+          Logger.warn(`Failed to rename file ${mdFile} to ${finalName}: ${renameError}`)
+          // 如果重命名失败，使用原文件
+          finalPath = originalMdPath
+          finalName = mdFile
+        }
+      }
+    } catch (error) {
+      Logger.warn(`Failed to read output directory ${outputPath}: ${error}`)
+      finalPath = path.join(outputPath, `${file.id}.md`)
+    }
+
+    return {
+      ...file,
+      name: finalName,
+      path: finalPath,
+      ext: '.md',
+      size: fs.existsSync(finalPath) ? fs.statSync(finalPath).size : 0
+    }
+  }
+
+  private async downloadAndExtractFile(zipUrl: string, file: FileMetadata): Promise<{ path: string }> {
+    const dirPath = this.storageDir
+
+    const zipPath = path.join(dirPath, `${file.id}.zip`)
+    const extractPath = path.join(dirPath, `${file.id}`)
+
+    Logger.info(`Downloading MinerU result to: ${zipPath}`)
+
+    try {
+      // 下载ZIP文件
+      const response = await axios.get(zipUrl, { responseType: 'arraybuffer' })
+      fs.writeFileSync(zipPath, response.data)
+      Logger.info(`Downloaded ZIP file: ${zipPath}`)
+
+      // 确保提取目录存在
+      if (!fs.existsSync(extractPath)) {
+        fs.mkdirSync(extractPath, { recursive: true })
+      }
+
+      // 解压文件
+      const zip = new AdmZip(zipPath)
+      zip.extractAllTo(extractPath, true)
+      Logger.info(`Extracted files to: ${extractPath}`)
+
+      // 删除临时ZIP文件
+      fs.unlinkSync(zipPath)
+
+      return { path: extractPath }
+    } catch (error: any) {
+      Logger.error(`Failed to download and extract file: ${error.message}`)
+      throw new Error(error.message)
+    }
+  }
+
+  private async uploadFile(file: FileMetadata): Promise<string> {
+    try {
+      // 步骤1: 获取上传URL
+      const { batchId, fileUrls } = await this.getBatchUploadUrls(file)
+      Logger.info(`Got upload URLs for batch: ${batchId}`)
+
+      console.log('batchId:', batchId, 'fileurls:', fileUrls)
+      // 步骤2: 上传文件到获取的URL
+      await this.putFileToUrl(file.path, fileUrls[0])
+      Logger.info(`File uploaded successfully: ${file.path}`)
+
+      return batchId
+    } catch (error: any) {
+      Logger.error(`Failed to upload file ${file.path}: ${error.message}`)
+      throw new Error(error.message)
+    }
+  }
+
+  private async getBatchUploadUrls(file: FileMetadata): Promise<{ batchId: string; fileUrls: string[] }> {
+    const endpoint = `${this.provider.apiHost}/api/v4/file-urls/batch`
+
+    const payload = {
+      language: 'auto',
+      enable_formula: true,
+      enable_table: true,
+      files: [
+        {
+          name: file.origin_name,
+          is_ocr: true,
+          data_id: file.id
+        }
+      ]
+    }
+
+    try {
+      const response = await fetch(endpoint, {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          Authorization: `Bearer ${this.provider.apiKey}`,
+          token: this.userId ?? '',
+          Accept: '*/*'
+        },
+        body: JSON.stringify(payload)
+      })
+
+      if (response.ok) {
+        const data: ApiResponse<BatchUploadResponse> = await response.json()
+        if (data.code === 0 && data.data) {
+          const { batch_id, file_urls } = data.data
+          return {
+            batchId: batch_id,
+            fileUrls: file_urls
+          }
+        } else {
+          throw new Error(`API returned error: ${data.msg || JSON.stringify(data)}`)
+        }
+      } else {
+        throw new Error(`HTTP ${response.status}: ${response.statusText}`)
+      }
+    } catch (error: any) {
+      Logger.error(`Failed to get batch upload URLs: ${error.message}`)
+      throw new Error(error.message)
+    }
+  }
+
+  private async putFileToUrl(filePath: string, uploadUrl: string): Promise<void> {
+    try {
+      const fileBuffer = await fs.promises.readFile(filePath)
+
+      const response = await fetch(uploadUrl, {
+        method: 'PUT',
+        body: fileBuffer,
+        headers: {
+          'Content-Type': 'application/pdf'
+        }
+        // headers: {
+        //   'Content-Length': fileBuffer.length.toString()
+        // }
+      })
+
+      if (!response.ok) {
+        // 克隆 response 以避免消费 body stream
+        const responseClone = response.clone()
+
+        try {
+          const responseBody = await responseClone.text()
+          const errorInfo = {
+            status: response.status,
+            statusText: response.statusText,
+            url: response.url,
+            type: response.type,
+            redirected: response.redirected,
+            headers: Object.fromEntries(response.headers.entries()),
+            body: responseBody
+          }
+
+          console.error('Response details:', errorInfo)
+          throw new Error(`Upload failed with status ${response.status}: ${responseBody}`)
+        } catch (parseError) {
+          throw new Error(`Upload failed with status ${response.status}. Could not parse response body.`)
+        }
+      }
+
+      Logger.info(`File uploaded successfully to: ${uploadUrl}`)
+    } catch (error: any) {
+      Logger.error(`Failed to upload file to URL ${uploadUrl}: ${error}`)
+      throw new Error(error.message)
+    }
+  }
+
+  private async getExtractResults(batchId: string): Promise<ExtractResultResponse> {
+    const endpoint = `${this.provider.apiHost}/api/v4/extract-results/batch/${batchId}`
+
+    try {
+      const response = await fetch(endpoint, {
+        method: 'GET',
+        headers: {
+          'Content-Type': 'application/json',
+          Authorization: `Bearer ${this.provider.apiKey}`,
+          token: this.userId ?? ''
+        }
+      })
+
+      if (response.ok) {
+        const data: ApiResponse<ExtractResultResponse> = await response.json()
+        if (data.code === 0 && data.data) {
+          return data.data
+        } else {
+          throw new Error(`API returned error: ${data.msg || JSON.stringify(data)}`)
+        }
+      } else {
+        throw new Error(`HTTP ${response.status}: ${response.statusText}`)
+      }
+    } catch (error: any) {
+      Logger.error(`Failed to get extract results for batch ${batchId}: ${error.message}`)
+      throw new Error(error.message)
+    }
+  }
+
+  private async waitForCompletion(
+    sourceId: string,
+    batchId: string,
+    fileName: string,
+    maxRetries: number = 60,
+    intervalMs: number = 5000
+  ): Promise<ExtractFileResult> {
+    let retries = 0
+
+    while (retries < maxRetries) {
+      try {
+        const result = await this.getExtractResults(batchId)
+
+        // 查找对应文件的处理结果
+        const fileResult = result.extract_result.find((item) => item.file_name === fileName)
+        if (!fileResult) {
+          throw new Error(`File ${fileName} not found in batch results`)
+        }
+
+        // 检查处理状态
+        if (fileResult.state === 'done' && fileResult.full_zip_url) {
+          Logger.info(`Processing completed for file: ${fileName}`)
+          return fileResult
+        } else if (fileResult.state === 'failed') {
+          throw new Error(`Processing failed for file: ${fileName}, error: ${fileResult.err_msg}`)
+        } else if (fileResult.state === 'running') {
+          // 发送进度更新
+          if (fileResult.extract_progress) {
+            const progress = Math.round(
+              (fileResult.extract_progress.extracted_pages / fileResult.extract_progress.total_pages) * 100
+            )
+            await this.sendPreprocessProgress(sourceId, progress)
+            Logger.info(`File ${fileName} processing progress: ${progress}%`)
+          } else {
+            // 如果没有具体进度信息，发送一个通用进度
+            await this.sendPreprocessProgress(sourceId, 50)
+            Logger.info(`File ${fileName} is still processing...`)
+          }
+        }
+      } catch (error) {
+        Logger.warn(`Failed to check status for batch ${batchId}, retry ${retries + 1}/${maxRetries}`)
+        if (retries === maxRetries - 1) {
+          throw error
+        }
+      }
+
+      retries++
+      await new Promise((resolve) => setTimeout(resolve, intervalMs))
+    }
+
+    throw new Error(`Processing timeout for batch: ${batchId}`)
+  }
+}

src/main/preprocess/MistralPreprocessProvider.ts

@@ -0,0 +1,187 @@
+import fs from 'node:fs'
+
+import { MistralClientManager } from '@main/services/MistralClientManager'
+import { MistralService } from '@main/services/remotefile/MistralService'
+import { Mistral } from '@mistralai/mistralai'
+import { DocumentURLChunk } from '@mistralai/mistralai/models/components/documenturlchunk'
+import { ImageURLChunk } from '@mistralai/mistralai/models/components/imageurlchunk'
+import { OCRResponse } from '@mistralai/mistralai/models/components/ocrresponse'
+import { FileMetadata, FileTypes, PreprocessProvider, Provider } from '@types'
+import Logger from 'electron-log'
+import path from 'path'
+
+import BasePreprocessProvider from './BasePreprocessProvider'
+
+type PreuploadResponse = DocumentURLChunk | ImageURLChunk
+
+export default class MistralPreprocessProvider extends BasePreprocessProvider {
+  private sdk: Mistral
+  private fileService: MistralService
+
+  constructor(provider: PreprocessProvider) {
+    super(provider)
+    const clientManager = MistralClientManager.getInstance()
+    const aiProvider: Provider = {
+      id: provider.id,
+      type: 'mistral',
+      name: provider.name,
+      apiKey: provider.apiKey!,
+      apiHost: provider.apiHost!,
+      models: []
+    }
+    clientManager.initializeClient(aiProvider)
+    this.sdk = clientManager.getClient()
+    this.fileService = new MistralService(aiProvider)
+  }
+
+  private async preupload(file: FileMetadata): Promise<PreuploadResponse> {
+    let document: PreuploadResponse
+    Logger.info(`preprocess preupload started for local file: ${file.path}`)
+
+    if (file.ext.toLowerCase() === '.pdf') {
+      const uploadResponse = await this.fileService.uploadFile(file)
+
+      if (uploadResponse.status === 'failed') {
+        Logger.error('File upload failed:', uploadResponse)
+        throw new Error('Failed to upload file: ' + uploadResponse.displayName)
+      }
+      await this.sendPreprocessProgress(file.id, 15)
+      const fileUrl = await this.sdk.files.getSignedUrl({
+        fileId: uploadResponse.fileId
+      })
+      Logger.info('Got signed URL:', fileUrl)
+      await this.sendPreprocessProgress(file.id, 20)
+      document = {
+        type: 'document_url',
+        documentUrl: fileUrl.url
+      }
+    } else {
+      const base64Image = Buffer.from(fs.readFileSync(file.path)).toString('base64')
+      document = {
+        type: 'image_url',
+        imageUrl: `data:image/png;base64,${base64Image}`
+      }
+    }
+
+    if (!document) {
+      throw new Error('Unsupported file type')
+    }
+    return document
+  }
+
+  public async parseFile(sourceId: string, file: FileMetadata): Promise<{ processedFile: FileMetadata }> {
+    try {
+      const document = await this.preupload(file)
+      const result = await this.sdk.ocr.process({
+        model: this.provider.model!,
+        document: document,
+        includeImageBase64: true
+      })
+      if (result) {
+        await this.sendPreprocessProgress(sourceId, 100)
+        const processedFile = this.convertFile(result, file)
+        return {
+          processedFile
+        }
+      } else {
+        throw new Error('preprocess processing failed: OCR response is empty')
+      }
+    } catch (error) {
+      throw new Error('preprocess processing failed: ' + error)
+    }
+  }
+
+  private convertFile(result: OCRResponse, file: FileMetadata): FileMetadata {
+    // 使用统一的存储路径：Data/Files/{file.id}/
+    const conversionId = file.id
+    const outputPath = path.join(this.storageDir, file.id)
+    // const outputPath = this.storageDir
+    const outputFileName = path.basename(file.path, path.extname(file.path))
+    fs.mkdirSync(outputPath, { recursive: true })
+
+    const markdownParts: string[] = []
+    let counter = 0
+
+    // Process each page
+    result.pages.forEach((page) => {
+      let pageMarkdown = page.markdown
+
+      // Process images from this page
+      page.images.forEach((image) => {
+        if (image.imageBase64) {
+          let imageFormat = 'jpeg' // default format
+          let imageBase64Data = image.imageBase64
+
+          // Check for data URL prefix more efficiently
+          const prefixEnd = image.imageBase64.indexOf(';base64,')
+          if (prefixEnd > 0) {
+            const prefix = image.imageBase64.substring(0, prefixEnd)
+            const formatIndex = prefix.indexOf('image/')
+            if (formatIndex >= 0) {
+              imageFormat = prefix.substring(formatIndex + 6)
+            }
+            imageBase64Data = image.imageBase64.substring(prefixEnd + 8)
+          }
+
+          const imageFileName = `img-${counter}.${imageFormat}`
+          const imagePath = path.join(outputPath, imageFileName)
+
+          // Save image file
+          try {
+            fs.writeFileSync(imagePath, Buffer.from(imageBase64Data, 'base64'))
+
+            // Update image reference in markdown
+            // Use relative path for better portability
+            const relativeImagePath = `./${imageFileName}`
+
+            // Find the start and end of the image markdown
+            const imgStart = pageMarkdown.indexOf(image.imageBase64)
+            if (imgStart >= 0) {
+              // Find the markdown image syntax around this base64
+              const mdStart = pageMarkdown.lastIndexOf('![', imgStart)
+              const mdEnd = pageMarkdown.indexOf(')', imgStart)
+
+              if (mdStart >= 0 && mdEnd >= 0) {
+                // Replace just this specific image reference
+                pageMarkdown =
+                  pageMarkdown.substring(0, mdStart) +
+                  `![Image ${counter}](${relativeImagePath})` +
+                  pageMarkdown.substring(mdEnd + 1)
+              }
+            }
+
+            counter++
+          } catch (error) {
+            Logger.error(`Failed to save image ${imageFileName}:`, error)
+          }
+        }
+      })
+
+      markdownParts.push(pageMarkdown)
+    })
+
+    // Combine all markdown content with double newlines for readability
+    const combinedMarkdown = markdownParts.join('\n\n')
+
+    // Write the markdown content to a file
+    const mdFileName = `${outputFileName}.md`
+    const mdFilePath = path.join(outputPath, mdFileName)
+    fs.writeFileSync(mdFilePath, combinedMarkdown)
+
+    return {
+      id: conversionId,
+      name: file.name.replace(/\.[^/.]+$/, '.md'),
+      origin_name: file.origin_name,
+      path: mdFilePath,
+      created_at: new Date().toISOString(),
+      type: FileTypes.DOCUMENT,
+      ext: '.md',
+      size: fs.statSync(mdFilePath).size,
+      count: 1
+    } as FileMetadata
+  }
+
+  public checkQuota(): Promise<number> {
+    throw new Error('Method not implemented.')
+  }
+}

src/main/preprocess/PreprocessProvider.ts

@@ -0,0 +1,30 @@
+import { FileMetadata, PreprocessProvider as Provider } from '@types'
+
+import BasePreprocessProvider from './BasePreprocessProvider'
+import PreprocessProviderFactory from './PreprocessProviderFactory'
+
+export default class PreprocessProvider {
+  private sdk: BasePreprocessProvider
+  constructor(provider: Provider, userId?: string) {
+    this.sdk = PreprocessProviderFactory.create(provider, userId)
+  }
+  public async parseFile(
+    sourceId: string,
+    file: FileMetadata
+  ): Promise<{ processedFile: FileMetadata; quota?: number }> {
+    return this.sdk.parseFile(sourceId, file)
+  }
+
+  public async checkQuota(): Promise<number> {
+    return this.sdk.checkQuota()
+  }
+
+  /**
+   * 检查文件是否已经被预处理过
+   * @param file 文件信息
+   * @returns 如果已处理返回处理后的文件信息，否则返回null
+   */
+  public async checkIfAlreadyProcessed(file: FileMetadata): Promise<FileMetadata | null> {
+    return this.sdk.checkIfAlreadyProcessed(file)
+  }
+}

src/main/preprocess/PreprocessProviderFactory.ts

@@ -0,0 +1,21 @@
+import { PreprocessProvider } from '@types'
+
+import BasePreprocessProvider from './BasePreprocessProvider'
+import DefaultPreprocessProvider from './DefaultPreprocessProvider'
+import Doc2xPreprocessProvider from './Doc2xPreprocessProvider'
+import MineruPreprocessProvider from './MineruPreprocessProvider'
+import MistralPreprocessProvider from './MistralPreprocessProvider'
+export default class PreprocessProviderFactory {
+  static create(provider: PreprocessProvider, userId?: string): BasePreprocessProvider {
+    switch (provider.id) {
+      case 'doc2x':
+        return new Doc2xPreprocessProvider(provider)
+      case 'mistral':
+        return new MistralPreprocessProvider(provider)
+      case 'mineru':
+        return new MineruPreprocessProvider(provider, userId)
+      default:
+        return new DefaultPreprocessProvider(provider)
+    }
+  }
+}

src/main/services/AppUpdater.ts

@@ -1,11 +1,13 @@
 import { isWin } from '@main/constant'
 import { locales } from '@main/utils/locales'
+import { generateUserAgent } from '@main/utils/systemInfo'
+import { FeedUrl, UpgradeChannel } from '@shared/config/constant'
 import { IpcChannel } from '@shared/IpcChannel'
-import { FeedUrl } from '@shared/config/constant'
-import { UpdateInfo } from 'builder-util-runtime'
+import { CancellationToken, UpdateInfo } from 'builder-util-runtime'
 import { app, BrowserWindow, dialog } from 'electron'
 import logger from 'electron-log'
-import { AppUpdater as _AppUpdater, autoUpdater } from 'electron-updater'
+import { AppUpdater as _AppUpdater, autoUpdater, NsisUpdater, UpdateCheckResult } from 'electron-updater'
+import path from 'path'
 
 import icon from '../../../build/icon.png?asset'
 import { configManager } from './ConfigManager'
@@ -13,6 +15,8 @@ import { configManager } from './ConfigManager'
 export default class AppUpdater {
   autoUpdater: _AppUpdater = autoUpdater
   private releaseInfo: UpdateInfo | undefined
+  private cancellationToken: CancellationToken = new CancellationToken()
+  private updateCheckResult: UpdateCheckResult | null = null
 
   constructor(mainWindow: BrowserWindow) {
     logger.transports.file.level = 'info'
@@ -21,9 +25,11 @@ export default class AppUpdater {
     autoUpdater.forceDevUpdateConfig = !app.isPackaged
     autoUpdater.autoDownload = configManager.getAutoUpdate()
     autoUpdater.autoInstallOnAppQuit = configManager.getAutoUpdate()
-    autoUpdater.setFeedURL(configManager.getFeedUrl())
+    autoUpdater.requestHeaders = {
+      ...autoUpdater.requestHeaders,
+      'User-Agent': generateUserAgent()
+    }
 
-    // 检测下载错误
     autoUpdater.on('error', (error) => {
       // 简单记录错误信息和时间戳
       logger.error('更新异常', {
@@ -56,17 +62,137 @@ export default class AppUpdater {
       logger.info('下载完成', releaseInfo)
     })
 
+    if (isWin) {
+      ;(autoUpdater as NsisUpdater).installDirectory = path.dirname(app.getPath('exe'))
+    }
+
     this.autoUpdater = autoUpdater
   }
 
+  private async _getPreReleaseVersionFromGithub(channel: UpgradeChannel) {
+    try {
+      logger.info('get pre release version from github', channel)
+      const responses = await fetch('https://api.github.com/repos/CherryHQ/cherry-studio/releases?per_page=8', {
+        headers: {
+          Accept: 'application/vnd.github+json',
+          'X-GitHub-Api-Version': '2022-11-28',
+          'Accept-Language': 'en-US,en;q=0.9'
+        }
+      })
+      const data = (await responses.json()) as GithubReleaseInfo[]
+      const release: GithubReleaseInfo | undefined = data.find((item: GithubReleaseInfo) => {
+        return item.prerelease && item.tag_name.includes(`-${channel}.`)
+      })
+
+      logger.info('release info', release)
+
+      if (!release) {
+        return null
+      }
+
+      logger.info('release info', release.tag_name)
+      return `https://github.com/CherryHQ/cherry-studio/releases/download/${release.tag_name}`
+    } catch (error) {
+      logger.error('Failed to get latest not draft version from github:', error)
+      return null
+    }
+  }
+
+  private async _getIpCountry() {
+    try {
+      // add timeout using AbortController
+      const controller = new AbortController()
+      const timeoutId = setTimeout(() => controller.abort(), 5000)
+
+      const ipinfo = await fetch('https://ipinfo.io/json', {
+        signal: controller.signal,
+        headers: {
+          'User-Agent':
+            'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36',
+          'Accept-Language': 'en-US,en;q=0.9'
+        }
+      })
+
+      clearTimeout(timeoutId)
+      const data = await ipinfo.json()
+      return data.country || 'CN'
+    } catch (error) {
+      logger.error('Failed to get ipinfo:', error)
+      return 'CN'
+    }
+  }
+
   public setAutoUpdate(isActive: boolean) {
     autoUpdater.autoDownload = isActive
     autoUpdater.autoInstallOnAppQuit = isActive
   }
 
-  public setFeedUrl(feedUrl: FeedUrl) {
-    autoUpdater.setFeedURL(feedUrl)
-    configManager.setFeedUrl(feedUrl)
+  private _getChannelByVersion(version: string) {
+    if (version.includes(`-${UpgradeChannel.BETA}.`)) {
+      return UpgradeChannel.BETA
+    }
+    if (version.includes(`-${UpgradeChannel.RC}.`)) {
+      return UpgradeChannel.RC
+    }
+    return UpgradeChannel.LATEST
+  }
+
+  private _getTestChannel() {
+    const currentChannel = this._getChannelByVersion(app.getVersion())
+    const savedChannel = configManager.getTestChannel()
+
+    if (currentChannel === UpgradeChannel.LATEST) {
+      return savedChannel || UpgradeChannel.RC
+    }
+
+    if (savedChannel === currentChannel) {
+      return savedChannel
+    }
+
+    // if the upgrade channel is not equal to the current channel, use the latest channel
+    return UpgradeChannel.LATEST
+  }
+
+  private async _setFeedUrl() {
+    const testPlan = configManager.getTestPlan()
+    if (testPlan) {
+      const channel = this._getTestChannel()
+
+      if (channel === UpgradeChannel.LATEST) {
+        this.autoUpdater.channel = UpgradeChannel.LATEST
+        this.autoUpdater.setFeedURL(FeedUrl.GITHUB_LATEST)
+        return
+      }
+
+      const preReleaseUrl = await this._getPreReleaseVersionFromGithub(channel)
+      if (preReleaseUrl) {
+        this.autoUpdater.setFeedURL(preReleaseUrl)
+        this.autoUpdater.channel = channel
+        return
+      }
+
+      // if no prerelease url, use lowest prerelease version to avoid error
+      this.autoUpdater.setFeedURL(FeedUrl.PRERELEASE_LOWEST)
+      this.autoUpdater.channel = UpgradeChannel.LATEST
+      return
+    }
+
+    this.autoUpdater.channel = UpgradeChannel.LATEST
+    this.autoUpdater.setFeedURL(FeedUrl.PRODUCTION)
+
+    const ipCountry = await this._getIpCountry()
+    logger.info('ipCountry', ipCountry)
+    if (ipCountry.toLowerCase() !== 'cn') {
+      this.autoUpdater.setFeedURL(FeedUrl.GITHUB_LATEST)
+    }
+  }
+
+  public cancelDownload() {
+    this.cancellationToken.cancel()
+    this.cancellationToken = new CancellationToken()
+    if (this.autoUpdater.autoDownload) {
+      this.updateCheckResult?.cancellationToken?.cancel()
+    }
   }
 
   public async checkForUpdates() {
@@ -77,17 +203,26 @@ export default class AppUpdater {
       }
     }
 
+    await this._setFeedUrl()
+
+    // disable downgrade after change the channel
+    this.autoUpdater.allowDowngrade = false
+
+    // github and gitcode don't support multiple range download
+    this.autoUpdater.disableDifferentialDownload = true
+
     try {
-      const update = await this.autoUpdater.checkForUpdates()
-      if (update?.isUpdateAvailable && !this.autoUpdater.autoDownload) {
+      this.updateCheckResult = await this.autoUpdater.checkForUpdates()
+      if (this.updateCheckResult?.isUpdateAvailable && !this.autoUpdater.autoDownload) {
         // 如果 autoDownload 为 false，则需要再调用下面的函数触发下
         // do not use await, because it will block the return of this function
-        this.autoUpdater.downloadUpdate()
+        logger.info('downloadUpdate manual by check for updates', this.cancellationToken)
+        this.autoUpdater.downloadUpdate(this.cancellationToken)
       }
 
       return {
         currentVersion: this.autoUpdater.currentVersion,
-        updateInfo: update?.updateInfo
+        updateInfo: this.updateCheckResult?.updateInfo
       }
     } catch (error) {
       logger.error('Failed to check for update:', error)
@@ -143,7 +278,11 @@ export default class AppUpdater {
     return releaseNotes.map((note) => note.note).join('\n')
   }
 }
-
+interface GithubReleaseInfo {
+  draft: boolean
+  prerelease: boolean
+  tag_name: string
+}
 interface ReleaseNoteInfo {
   readonly version: string
   readonly note: string | null

src/main/services/BackupManager.ts

@@ -9,6 +9,7 @@ import StreamZip from 'node-stream-zip'
 import * as path from 'path'
 import { CreateDirectoryOptions, FileStat } from 'webdav'
 
+import { getDataPath } from '../utils'
 import WebDav from './WebDav'
 import { windowService } from './WindowService'
 
@@ -253,7 +254,7 @@ class BackupManager {
       Logger.log('[backup] step 3: restore Data directory')
       // 恢复 Data 目录
       const sourcePath = path.join(this.tempDir, 'Data')
-      const destPath = path.join(app.getPath('userData'), 'Data')
+      const destPath = getDataPath()
 
       const dataExists = await fs.pathExists(sourcePath)
       const dataFiles = dataExists ? await fs.readdir(sourcePath) : []

src/main/services/ConfigManager.ts

@@ -1,4 +1,4 @@
-import { defaultLanguage, FeedUrl, ZOOM_SHORTCUTS } from '@shared/config/constant'
+import { defaultLanguage, UpgradeChannel, ZOOM_SHORTCUTS } from '@shared/config/constant'
 import { LanguageVarious, Shortcut, ThemeMode } from '@types'
 import { app } from 'electron'
 import Store from 'electron-store'
@@ -16,7 +16,8 @@ export enum ConfigKeys {
   ClickTrayToShowQuickAssistant = 'clickTrayToShowQuickAssistant',
   EnableQuickAssistant = 'enableQuickAssistant',
   AutoUpdate = 'autoUpdate',
-  FeedUrl = 'feedUrl',
+  TestPlan = 'testPlan',
+  TestChannel = 'testChannel',
   EnableDataCollection = 'enableDataCollection',
   SelectionAssistantEnabled = 'selectionAssistantEnabled',
   SelectionAssistantTriggerMode = 'selectionAssistantTriggerMode',
@@ -142,12 +143,20 @@ export class ConfigManager {
     this.set(ConfigKeys.AutoUpdate, value)
   }
 
-  getFeedUrl(): string {
-    return this.get<string>(ConfigKeys.FeedUrl, FeedUrl.PRODUCTION)
+  getTestPlan(): boolean {
+    return this.get<boolean>(ConfigKeys.TestPlan, false)
   }
 
-  setFeedUrl(value: FeedUrl) {
-    this.set(ConfigKeys.FeedUrl, value)
+  setTestPlan(value: boolean) {
+    this.set(ConfigKeys.TestPlan, value)
+  }
+
+  getTestChannel(): UpgradeChannel {
+    return this.get<UpgradeChannel>(ConfigKeys.TestChannel)
+  }
+
+  setTestChannel(value: UpgradeChannel) {
+    this.set(ConfigKeys.TestChannel, value)
   }
 
   getEnableDataCollection(): boolean {

src/main/services/ContextMenu.ts

@@ -4,18 +4,29 @@ import { locales } from '../utils/locales'
 import { configManager } from './ConfigManager'
 
 class ContextMenu {
-  public contextMenu(w: Electron.BrowserWindow) {
-    w.webContents.on('context-menu', (_event, properties) => {
+  public contextMenu(w: Electron.WebContents) {
+    w.on('context-menu', (_event, properties) => {
       const template: MenuItemConstructorOptions[] = this.createEditMenuItems(properties)
       const filtered = template.filter((item) => item.visible !== false)
       if (filtered.length > 0) {
-        const menu = Menu.buildFromTemplate([...filtered, ...this.createInspectMenuItems(w)])
+        let template = [...filtered, ...this.createInspectMenuItems(w)]
+        const dictionarySuggestions = this.createDictionarySuggestions(properties, w)
+        if (dictionarySuggestions.length > 0) {
+          template = [
+            ...dictionarySuggestions,
+            { type: 'separator' },
+            this.createSpellCheckMenuItem(properties, w),
+            { type: 'separator' },
+            ...template
+          ]
+        }
+        const menu = Menu.buildFromTemplate(template)
         menu.popup()
       }
     })
   }
 
-  private createInspectMenuItems(w: Electron.BrowserWindow): MenuItemConstructorOptions[] {
+  private createInspectMenuItems(w: Electron.WebContents): MenuItemConstructorOptions[] {
     const locale = locales[configManager.getLanguage()]
     const { common } = locale.translation
     const template: MenuItemConstructorOptions[] = [
@@ -23,7 +34,7 @@ class ContextMenu {
         id: 'inspect',
         label: common.inspect,
         click: () => {
-          w.webContents.toggleDevTools()
+          w.toggleDevTools()
         },
         enabled: true
       }
@@ -72,6 +83,53 @@ class ContextMenu {
 
     return template
   }
+
+  private createSpellCheckMenuItem(
+    properties: Electron.ContextMenuParams,
+    w: Electron.WebContents
+  ): MenuItemConstructorOptions {
+    const hasText = properties.selectionText.length > 0
+
+    return {
+      id: 'learnSpelling',
+      label: '&Learn Spelling',
+      visible: Boolean(properties.isEditable && hasText && properties.misspelledWord),
+      click: () => {
+        w.session.addWordToSpellCheckerDictionary(properties.misspelledWord)
+      }
+    }
+  }
+
+  private createDictionarySuggestions(
+    properties: Electron.ContextMenuParams,
+    w: Electron.WebContents
+  ): MenuItemConstructorOptions[] {
+    const hasText = properties.selectionText.length > 0
+
+    if (!hasText || !properties.misspelledWord) {
+      return []
+    }
+
+    if (properties.dictionarySuggestions.length === 0) {
+      return [
+        {
+          id: 'dictionarySuggestions',
+          label: 'No Guesses Found',
+          visible: true,
+          enabled: false
+        }
+      ]
+    }
+
+    return properties.dictionarySuggestions.map((suggestion) => ({
+      id: 'dictionarySuggestions',
+      label: suggestion,
+      visible: Boolean(properties.isEditable && hasText && properties.misspelledWord),
+      click: (menuItem: Electron.MenuItem) => {
+        w.replaceMisspelling(menuItem.label)
+      }
+    }))
+  }
 }
 
 export const contextMenu = new ContextMenu()

src/main/services/FileStorage.ts

@@ -1,6 +1,6 @@
 import { getFilesDir, getFileType, getTempDir } from '@main/utils/file'
 import { documentExts, imageExts, MB } from '@shared/config/constant'
-import { FileType } from '@types'
+import { FileMetadata } from '@types'
 import * as crypto from 'crypto'
 import {
   dialog,
@@ -15,9 +15,11 @@ import * as fs from 'fs'
 import { writeFileSync } from 'fs'
 import { readFile } from 'fs/promises'
 import officeParser from 'officeparser'
+import { getDocument } from 'officeparser/pdfjs-dist-build/pdf.js'
 import * as path from 'path'
 import { chdir } from 'process'
 import { v4 as uuidv4 } from 'uuid'
+import WordExtractor from 'word-extractor'
 
 class FileStorage {
   private storageDir = getFilesDir()
@@ -51,8 +53,9 @@ class FileStorage {
     })
   }
 
-  findDuplicateFile = async (filePath: string): Promise<FileType | null> => {
+  findDuplicateFile = async (filePath: string): Promise<FileMetadata | null> => {
     const stats = fs.statSync(filePath)
+    console.log('stats', stats, filePath)
     const fileSize = stats.size
 
     const files = await fs.promises.readdir(this.storageDir)
@@ -90,7 +93,7 @@ class FileStorage {
   public selectFile = async (
     _: Electron.IpcMainInvokeEvent,
     options?: OpenDialogOptions
-  ): Promise<FileType[] | null> => {
+  ): Promise<FileMetadata[] | null> => {
     const defaultOptions: OpenDialogOptions = {
       properties: ['openFile']
     }
@@ -149,7 +152,7 @@ class FileStorage {
     }
   }
 
-  public uploadFile = async (_: Electron.IpcMainInvokeEvent, file: FileType): Promise<FileType> => {
+  public uploadFile = async (_: Electron.IpcMainInvokeEvent, file: FileMetadata): Promise<FileMetadata> => {
     const duplicateFile = await this.findDuplicateFile(file.path)
 
     if (duplicateFile) {
@@ -173,7 +176,7 @@ class FileStorage {
     const stats = await fs.promises.stat(destPath)
     const fileType = getFileType(ext)
 
-    const fileMetadata: FileType = {
+    const fileMetadata: FileMetadata = {
       id: uuid,
       origin_name,
       name: uuid + ext,
@@ -188,7 +191,7 @@ class FileStorage {
     return fileMetadata
   }
 
-  public getFile = async (_: Electron.IpcMainInvokeEvent, filePath: string): Promise<FileType | null> => {
+  public getFile = async (_: Electron.IpcMainInvokeEvent, filePath: string): Promise<FileMetadata | null> => {
     if (!fs.existsSync(filePath)) {
       return null
     }
@@ -197,7 +200,7 @@ class FileStorage {
     const ext = path.extname(filePath)
     const fileType = getFileType(ext)
 
-    const fileInfo: FileType = {
+    const fileInfo: FileMetadata = {
       id: uuidv4(),
       origin_name: path.basename(filePath),
       name: path.basename(filePath),
@@ -213,16 +216,36 @@ class FileStorage {
   }
 
   public deleteFile = async (_: Electron.IpcMainInvokeEvent, id: string): Promise<void> => {
+    if (!fs.existsSync(path.join(this.storageDir, id))) {
+      return
+    }
     await fs.promises.unlink(path.join(this.storageDir, id))
   }
 
+  public deleteDir = async (_: Electron.IpcMainInvokeEvent, id: string): Promise<void> => {
+    if (!fs.existsSync(path.join(this.storageDir, id))) {
+      return
+    }
+    await fs.promises.rm(path.join(this.storageDir, id), { recursive: true })
+  }
+
   public readFile = async (_: Electron.IpcMainInvokeEvent, id: string): Promise<string> => {
     const filePath = path.join(this.storageDir, id)
 
-    if (documentExts.includes(path.extname(filePath))) {
+    const fileExtension = path.extname(filePath)
+
+    if (documentExts.includes(fileExtension)) {
       const originalCwd = process.cwd()
       try {
         chdir(this.tempDir)
+
+        if (fileExtension === '.doc') {
+          const extractor = new WordExtractor()
+          const extracted = await extractor.extract(filePath)
+          chdir(originalCwd)
+          return extracted.getBody()
+        }
+
         const data = await officeParser.parseOfficeAsync(filePath)
         chdir(originalCwd)
         return data
@@ -240,8 +263,8 @@ class FileStorage {
     if (!fs.existsSync(this.tempDir)) {
       fs.mkdirSync(this.tempDir, { recursive: true })
     }
-    const tempFilePath = path.join(this.tempDir, `temp_file_${uuidv4()}_${fileName}`)
-    return tempFilePath
+
+    return path.join(this.tempDir, `temp_file_${uuidv4()}_${fileName}`)
   }
 
   public writeFile = async (
@@ -268,7 +291,7 @@ class FileStorage {
     }
   }
 
-  public saveBase64Image = async (_: Electron.IpcMainInvokeEvent, base64Data: string): Promise<FileType> => {
+  public saveBase64Image = async (_: Electron.IpcMainInvokeEvent, base64Data: string): Promise<FileMetadata> => {
     try {
       if (!base64Data) {
         throw new Error('Base64 data is required')
@@ -294,7 +317,7 @@ class FileStorage {
 
       await fs.promises.writeFile(destPath, buffer)
 
-      const fileMetadata: FileType = {
+      const fileMetadata: FileMetadata = {
         id: uuid,
         origin_name: uuid + ext,
         name: uuid + ext,
@@ -321,6 +344,16 @@ class FileStorage {
     return { data: base64, mime }
   }
 
+  public pdfPageCount = async (_: Electron.IpcMainInvokeEvent, id: string): Promise<number> => {
+    const filePath = path.join(this.storageDir, id)
+    const buffer = await fs.promises.readFile(filePath)
+
+    const doc = await getDocument({ data: buffer }).promise
+    const pages = doc.numPages
+    await doc.destroy()
+    return pages
+  }
+
   public binaryImage = async (_: Electron.IpcMainInvokeEvent, id: string): Promise<{ data: Buffer; mime: string }> => {
     const filePath = path.join(this.storageDir, id)
     const data = await fs.promises.readFile(filePath)
@@ -341,7 +374,7 @@ class FileStorage {
   public open = async (
     _: Electron.IpcMainInvokeEvent,
     options: OpenDialogOptions
-  ): Promise<{ fileName: string; filePath: string; content: Buffer } | null> => {
+  ): Promise<{ fileName: string; filePath: string; content?: Buffer; size: number } | null> => {
     try {
       const result: OpenDialogReturnValue = await dialog.showOpenDialog({
         title: '打开文件',
@@ -353,8 +386,16 @@ class FileStorage {
       if (!result.canceled && result.filePaths.length > 0) {
         const filePath = result.filePaths[0]
         const fileName = filePath.split('/').pop() || ''
-        const content = await readFile(filePath)
-        return { fileName, filePath, content }
+        const stats = await fs.promises.stat(filePath)
+
+        // If the file is less than 2GB, read the content
+        if (stats.size < 2 * 1024 * 1024 * 1024) {
+          const content = await readFile(filePath)
+          return { fileName, filePath, content, size: stats.size }
+        }
+
+        // For large files, only return file information, do not read content
+        return { fileName, filePath, size: stats.size }
       }
 
       return null
@@ -435,7 +476,7 @@ class FileStorage {
     _: Electron.IpcMainInvokeEvent,
     url: string,
     isUseContentType?: boolean
-  ): Promise<FileType> => {
+  ): Promise<FileMetadata> => {
     try {
       const response = await fetch(url)
       if (!response.ok) {
@@ -477,7 +518,7 @@ class FileStorage {
       const stats = await fs.promises.stat(destPath)
       const fileType = getFileType(ext)
 
-      const fileMetadata: FileType = {
+      const fileMetadata: FileMetadata = {
         id: uuid,
         origin_name: filename,
         name: uuid + ext,

src/main/services/KnowledgeService.ts

@@ -16,21 +16,24 @@
 import * as fs from 'node:fs'
 import path from 'node:path'
 
-import { RAGApplication, RAGApplicationBuilder, TextLoader } from '@cherrystudio/embedjs'
+import { RAGApplication, RAGApplicationBuilder } from '@cherrystudio/embedjs'
 import type { ExtractChunkData } from '@cherrystudio/embedjs-interfaces'
 import { LibSqlDb } from '@cherrystudio/embedjs-libsql'
 import { SitemapLoader } from '@cherrystudio/embedjs-loader-sitemap'
 import { WebLoader } from '@cherrystudio/embedjs-loader-web'
-import Embeddings from '@main/embeddings/Embeddings'
-import { addFileLoader } from '@main/loader'
-import Reranker from '@main/reranker/Reranker'
+import Embeddings from '@main/knowledage/embeddings/Embeddings'
+import { addFileLoader } from '@main/knowledage/loader'
+import { NoteLoader } from '@main/knowledage/loader/noteLoader'
+import Reranker from '@main/knowledage/reranker/Reranker'
+import OcrProvider from '@main/ocr/OcrProvider'
+import PreprocessProvider from '@main/preprocess/PreprocessProvider'
 import { windowService } from '@main/services/WindowService'
+import { getDataPath } from '@main/utils'
 import { getAllFiles } from '@main/utils/file'
 import { MB } from '@shared/config/constant'
 import type { LoaderReturn } from '@shared/config/types'
 import { IpcChannel } from '@shared/IpcChannel'
-import { FileType, KnowledgeBaseParams, KnowledgeItem } from '@types'
-import { app } from 'electron'
+import { FileMetadata, KnowledgeBaseParams, KnowledgeItem } from '@types'
 import Logger from 'electron-log'
 import { v4 as uuidv4 } from 'uuid'
 
@@ -38,12 +41,14 @@ export interface KnowledgeBaseAddItemOptions {
   base: KnowledgeBaseParams
   item: KnowledgeItem
   forceReload?: boolean
+  userId?: string
 }
 
 interface KnowledgeBaseAddItemOptionsNonNullableAttribute {
   base: KnowledgeBaseParams
   item: KnowledgeItem
   forceReload: boolean
+  userId: string
 }
 
 interface EvaluateTaskWorkload {
@@ -88,14 +93,20 @@ const loaderTaskIntoOfSet = (loaderTask: LoaderTask): LoaderTaskOfSet => {
 }
 
 class KnowledgeService {
-  private storageDir = path.join(app.getPath('userData'), 'Data', 'KnowledgeBase')
+  private storageDir = path.join(getDataPath(), 'KnowledgeBase')
   // Byte based
   private workload = 0
   private processingItemCount = 0
   private knowledgeItemProcessingQueueMappingPromise: Map<LoaderTaskOfSet, () => void> = new Map()
   private static MAXIMUM_WORKLOAD = 80 * MB
   private static MAXIMUM_PROCESSING_ITEM_COUNT = 30
-  private static ERROR_LOADER_RETURN: LoaderReturn = { entriesAdded: 0, uniqueId: '', uniqueIds: [''], loaderType: '' }
+  private static ERROR_LOADER_RETURN: LoaderReturn = {
+    entriesAdded: 0,
+    uniqueId: '',
+    uniqueIds: [''],
+    loaderType: '',
+    status: 'failed'
+  }
 
   constructor() {
     this.initStorageDir()
@@ -110,13 +121,21 @@ class KnowledgeService {
   private getRagApplication = async ({
     id,
     model,
+    provider,
     apiKey,
     apiVersion,
     baseURL,
     dimensions
   }: KnowledgeBaseParams): Promise<RAGApplication> => {
     let ragApplication: RAGApplication
-    const embeddings = new Embeddings({ model, apiKey, apiVersion, baseURL, dimensions } as KnowledgeBaseParams)
+    const embeddings = new Embeddings({
+      model,
+      provider,
+      apiKey,
+      apiVersion,
+      baseURL,
+      dimensions
+    } as KnowledgeBaseParams)
     try {
       ragApplication = await new RAGApplicationBuilder()
         .setModel('NO_MODEL')
@@ -135,12 +154,13 @@ class KnowledgeService {
     this.getRagApplication(base)
   }
 
-  public reset = async (_: Electron.IpcMainInvokeEvent, { base }: { base: KnowledgeBaseParams }): Promise<void> => {
+  public reset = async (_: Electron.IpcMainInvokeEvent, base: KnowledgeBaseParams): Promise<void> => {
     const ragApplication = await this.getRagApplication(base)
     await ragApplication.reset()
   }
 
   public delete = async (_: Electron.IpcMainInvokeEvent, id: string): Promise<void> => {
+    console.log('id', id)
     const dbPath = path.join(this.storageDir, id)
     if (fs.existsSync(dbPath)) {
       fs.rmSync(dbPath, { recursive: true })
@@ -153,28 +173,49 @@ class KnowledgeService {
       this.workload >= KnowledgeService.MAXIMUM_WORKLOAD
     )
   }
-
   private fileTask(
     ragApplication: RAGApplication,
     options: KnowledgeBaseAddItemOptionsNonNullableAttribute
   ): LoaderTask {
-    const { base, item, forceReload } = options
-    const file = item.content as FileType
+    const { base, item, forceReload, userId } = options
+    const file = item.content as FileMetadata
 
     const loaderTask: LoaderTask = {
       loaderTasks: [
         {
           state: LoaderTaskItemState.PENDING,
-          task: () =>
-            addFileLoader(ragApplication, file, base, forceReload)
-              .then((result) => {
-                loaderTask.loaderDoneReturn = result
-                return result
-              })
-              .catch((err) => {
-                Logger.error(err)
-                return KnowledgeService.ERROR_LOADER_RETURN
-              }),
+          task: async () => {
+            try {
+              // 添加预处理逻辑
+              const fileToProcess: FileMetadata = await this.preprocessing(file, base, item, userId)
+
+              // 使用处理后的文件进行加载
+              return addFileLoader(ragApplication, fileToProcess, base, forceReload)
+                .then((result) => {
+                  loaderTask.loaderDoneReturn = result
+                  return result
+                })
+                .catch((e) => {
+                  Logger.error(`Error in addFileLoader for ${file.name}: ${e}`)
+                  const errorResult: LoaderReturn = {
+                    ...KnowledgeService.ERROR_LOADER_RETURN,
+                    message: e.message,
+                    messageSource: 'embedding'
+                  }
+                  loaderTask.loaderDoneReturn = errorResult
+                  return errorResult
+                })
+            } catch (e: any) {
+              Logger.error(`Preprocessing failed for ${file.name}: ${e}`)
+              const errorResult: LoaderReturn = {
+                ...KnowledgeService.ERROR_LOADER_RETURN,
+                message: e.message,
+                messageSource: 'preprocess'
+              }
+              loaderTask.loaderDoneReturn = errorResult
+              return errorResult
+            }
+          },
           evaluateTaskWorkload: { workload: file.size }
         }
       ],
@@ -183,7 +224,6 @@ class KnowledgeService {
 
     return loaderTask
   }
-
   private directoryTask(
     ragApplication: RAGApplication,
     options: KnowledgeBaseAddItemOptionsNonNullableAttribute
@@ -223,7 +263,11 @@ class KnowledgeService {
             })
             .catch((err) => {
               Logger.error(err)
-              return KnowledgeService.ERROR_LOADER_RETURN
+              return {
+                ...KnowledgeService.ERROR_LOADER_RETURN,
+                message: `Failed to add dir loader: ${err.message}`,
+                messageSource: 'embedding'
+              }
             }),
         evaluateTaskWorkload: { workload: file.size }
       })
@@ -269,7 +313,11 @@ class KnowledgeService {
               })
               .catch((err) => {
                 Logger.error(err)
-                return KnowledgeService.ERROR_LOADER_RETURN
+                return {
+                  ...KnowledgeService.ERROR_LOADER_RETURN,
+                  message: `Failed to add url loader: ${err.message}`,
+                  messageSource: 'embedding'
+                }
               })
           },
           evaluateTaskWorkload: { workload: 2 * MB }
@@ -309,7 +357,11 @@ class KnowledgeService {
               })
               .catch((err) => {
                 Logger.error(err)
-                return KnowledgeService.ERROR_LOADER_RETURN
+                return {
+                  ...KnowledgeService.ERROR_LOADER_RETURN,
+                  message: `Failed to add sitemap loader: ${err.message}`,
+                  messageSource: 'embedding'
+                }
               }),
           evaluateTaskWorkload: { workload: 20 * MB }
         }
@@ -325,6 +377,7 @@ class KnowledgeService {
   ): LoaderTask {
     const { base, item, forceReload } = options
     const content = item.content as string
+    const sourceUrl = (item as any).sourceUrl
 
     const encoder = new TextEncoder()
     const contentBytes = encoder.encode(content)
@@ -334,7 +387,12 @@ class KnowledgeService {
           state: LoaderTaskItemState.PENDING,
           task: () => {
             const loaderReturn = ragApplication.addLoader(
-              new TextLoader({ text: content, chunkSize: base.chunkSize, chunkOverlap: base.chunkOverlap }),
+              new NoteLoader({
+                text: content,
+                sourceUrl,
+                chunkSize: base.chunkSize,
+                chunkOverlap: base.chunkOverlap
+              }),
               forceReload
             ) as Promise<LoaderReturn>
 
@@ -349,7 +407,11 @@ class KnowledgeService {
               })
               .catch((err) => {
                 Logger.error(err)
-                return KnowledgeService.ERROR_LOADER_RETURN
+                return {
+                  ...KnowledgeService.ERROR_LOADER_RETURN,
+                  message: `Failed to add note loader: ${err.message}`,
+                  messageSource: 'embedding'
+                }
               })
           },
           evaluateTaskWorkload: { workload: contentBytes.length }
@@ -415,10 +477,10 @@ class KnowledgeService {
     })
   }
 
-  public add = (_: Electron.IpcMainInvokeEvent, options: KnowledgeBaseAddItemOptions): Promise<LoaderReturn> => {
+  public add = async (_: Electron.IpcMainInvokeEvent, options: KnowledgeBaseAddItemOptions): Promise<LoaderReturn> => {
     return new Promise((resolve) => {
-      const { base, item, forceReload = false } = options
-      const optionsNonNullableAttribute = { base, item, forceReload }
+      const { base, item, forceReload = false, userId = '' } = options
+      const optionsNonNullableAttribute = { base, item, forceReload, userId }
       this.getRagApplication(base)
         .then((ragApplication) => {
           const task = (() => {
@@ -444,12 +506,20 @@ class KnowledgeService {
             })
             this.processingQueueHandle()
           } else {
-            resolve(KnowledgeService.ERROR_LOADER_RETURN)
+            resolve({
+              ...KnowledgeService.ERROR_LOADER_RETURN,
+              message: 'Unsupported item type',
+              messageSource: 'embedding'
+            })
           }
         })
         .catch((err) => {
           Logger.error(err)
-          resolve(KnowledgeService.ERROR_LOADER_RETURN)
+          resolve({
+            ...KnowledgeService.ERROR_LOADER_RETURN,
+            message: `Failed to add item: ${err.message}`,
+            messageSource: 'embedding'
+          })
         })
     })
   }
@@ -482,6 +552,69 @@ class KnowledgeService {
     }
     return await new Reranker(base).rerank(search, results)
   }
+
+  public getStorageDir = (): string => {
+    return this.storageDir
+  }
+
+  private preprocessing = async (
+    file: FileMetadata,
+    base: KnowledgeBaseParams,
+    item: KnowledgeItem,
+    userId: string
+  ): Promise<FileMetadata> => {
+    let fileToProcess: FileMetadata = file
+    if (base.preprocessOrOcrProvider && file.ext.toLowerCase() === '.pdf') {
+      try {
+        let provider: PreprocessProvider | OcrProvider
+        if (base.preprocessOrOcrProvider.type === 'preprocess') {
+          provider = new PreprocessProvider(base.preprocessOrOcrProvider.provider, userId)
+        } else {
+          provider = new OcrProvider(base.preprocessOrOcrProvider.provider)
+        }
+        // 首先检查文件是否已经被预处理过
+        const alreadyProcessed = await provider.checkIfAlreadyProcessed(file)
+        if (alreadyProcessed) {
+          Logger.info(`File already preprocess processed, using cached result: ${file.path}`)
+          return alreadyProcessed
+        }
+
+        // 执行预处理
+        Logger.info(`Starting preprocess processing for scanned PDF: ${file.path}`)
+        const { processedFile, quota } = await provider.parseFile(item.id, file)
+        fileToProcess = processedFile
+        const mainWindow = windowService.getMainWindow()
+        mainWindow?.webContents.send('file-preprocess-finished', {
+          itemId: item.id,
+          quota: quota
+        })
+      } catch (err) {
+        Logger.error(`Preprocess processing failed: ${err}`)
+        // 如果预处理失败，使用原始文件
+        // fileToProcess = file
+        throw new Error(`Preprocess processing failed: ${err}`)
+      }
+    }
+
+    return fileToProcess
+  }
+
+  public checkQuota = async (
+    _: Electron.IpcMainInvokeEvent,
+    base: KnowledgeBaseParams,
+    userId: string
+  ): Promise<number> => {
+    try {
+      if (base.preprocessOrOcrProvider && base.preprocessOrOcrProvider.type === 'preprocess') {
+        const provider = new PreprocessProvider(base.preprocessOrOcrProvider.provider, userId)
+        return await provider.checkQuota()
+      }
+      throw new Error('No preprocess provider configured')
+    } catch (err) {
+      Logger.error(`Failed to check quota: ${err}`)
+      throw new Error(`Failed to check quota: ${err}`)
+    }
+  }
 }
 
 export default new KnowledgeService()

src/main/services/MistralClientManager.ts

@@ -0,0 +1,33 @@
+import { Mistral } from '@mistralai/mistralai'
+import { Provider } from '@types'
+
+export class MistralClientManager {
+  private static instance: MistralClientManager
+  private client: Mistral | null = null
+
+  // eslint-disable-next-line @typescript-eslint/no-empty-function
+  private constructor() {}
+
+  public static getInstance(): MistralClientManager {
+    if (!MistralClientManager.instance) {
+      MistralClientManager.instance = new MistralClientManager()
+    }
+    return MistralClientManager.instance
+  }
+
+  public initializeClient(provider: Provider): void {
+    if (!this.client) {
+      this.client = new Mistral({
+        apiKey: provider.apiKey,
+        serverURL: provider.apiHost
+      })
+    }
+  }
+
+  public getClient(): Mistral {
+    if (!this.client) {
+      throw new Error('Mistral client not initialized. Call initializeClient first.')
+    }
+    return this.client
+  }
+}

src/main/services/ProtocolClient.ts

@@ -19,7 +19,7 @@ export function registerProtocolClient(app: Electron.App) {
     }
   }
 
-  app.setAsDefaultProtocolClient('cherrystudio')
+  app.setAsDefaultProtocolClient(CHERRY_STUDIO_PROTOCOL)
 }
 
 export function handleProtocolUrl(url: string) {

src/main/services/PythonService.ts

@@ -0,0 +1,102 @@
+import { randomUUID } from 'node:crypto'
+
+import { BrowserWindow, ipcMain } from 'electron'
+
+interface PythonExecutionRequest {
+  id: string
+  script: string
+  context: Record<string, any>
+  timeout: number
+}
+
+interface PythonExecutionResponse {
+  id: string
+  result?: string
+  error?: string
+}
+
+/**
+ * Service for executing Python code by communicating with the PyodideService in the renderer process
+ */
+export class PythonService {
+  private static instance: PythonService | null = null
+  private mainWindow: BrowserWindow | null = null
+  private pendingRequests = new Map<string, { resolve: (value: string) => void; reject: (error: Error) => void }>()
+
+  private constructor() {
+    // Private constructor for singleton pattern
+    this.setupIpcHandlers()
+  }
+
+  public static getInstance(): PythonService {
+    if (!PythonService.instance) {
+      PythonService.instance = new PythonService()
+    }
+    return PythonService.instance
+  }
+
+  private setupIpcHandlers() {
+    // Handle responses from renderer
+    ipcMain.on('python-execution-response', (_, response: PythonExecutionResponse) => {
+      const request = this.pendingRequests.get(response.id)
+      if (request) {
+        this.pendingRequests.delete(response.id)
+        if (response.error) {
+          request.reject(new Error(response.error))
+        } else {
+          request.resolve(response.result || '')
+        }
+      }
+    })
+  }
+
+  public setMainWindow(mainWindow: BrowserWindow) {
+    this.mainWindow = mainWindow
+  }
+
+  /**
+   * Execute Python code by sending request to renderer PyodideService
+   */
+  public async executeScript(
+    script: string,
+    context: Record<string, any> = {},
+    timeout: number = 60000
+  ): Promise<string> {
+    if (!this.mainWindow) {
+      throw new Error('Main window not set in PythonService')
+    }
+
+    return new Promise((resolve, reject) => {
+      const requestId = randomUUID()
+
+      // Store the request
+      this.pendingRequests.set(requestId, { resolve, reject })
+
+      // Set up timeout
+      const timeoutId = setTimeout(() => {
+        this.pendingRequests.delete(requestId)
+        reject(new Error('Python execution timed out'))
+      }, timeout + 5000) // Add 5s buffer for IPC communication
+
+      // Update resolve/reject to clear timeout
+      const originalResolve = resolve
+      const originalReject = reject
+      this.pendingRequests.set(requestId, {
+        resolve: (value: string) => {
+          clearTimeout(timeoutId)
+          originalResolve(value)
+        },
+        reject: (error: Error) => {
+          clearTimeout(timeoutId)
+          originalReject(error)
+        }
+      })
+
+      // Send request to renderer
+      const request: PythonExecutionRequest = { id: requestId, script, context, timeout }
+      this.mainWindow?.webContents.send('python-execution-request', request)
+    })
+  }
+}
+
+export const pythonService = PythonService.getInstance()

src/main/services/SelectionService.ts

@@ -1,7 +1,7 @@
 import { SELECTION_FINETUNED_LIST, SELECTION_PREDEFINED_BLACKLIST } from '@main/configs/SelectionConfig'
-import { isDev, isWin } from '@main/constant'
+import { isDev, isMac, isWin } from '@main/constant'
 import { IpcChannel } from '@shared/IpcChannel'
-import { BrowserWindow, ipcMain, screen } from 'electron'
+import { BrowserWindow, ipcMain, screen, systemPreferences } from 'electron'
 import Logger from 'electron-log'
 import { join } from 'path'
 import type {
@@ -14,10 +14,14 @@ import type {
 
 import type { ActionItem } from '../../renderer/src/types/selectionTypes'
 import { ConfigKeys, configManager } from './ConfigManager'
+import storeSyncService from './StoreSyncService'
+
+const isSupportedOS = isWin || isMac
 
 let SelectionHook: SelectionHookConstructor | null = null
 try {
-  if (isWin) {
+  //since selection-hook v1.0.0, it supports macOS
+  if (isSupportedOS) {
     SelectionHook = require('selection-hook')
   }
 } catch (error) {
@@ -39,7 +43,8 @@ type RelativeOrientation =
 
 enum TriggerMode {
   Selected = 'selected',
-  Ctrlkey = 'ctrlkey'
+  Ctrlkey = 'ctrlkey',
+  Shortcut = 'shortcut'
 }
 
 /** SelectionService is a singleton class that manages the selection hook and the toolbar window
@@ -116,7 +121,7 @@ export class SelectionService {
   }
 
   public static getInstance(): SelectionService | null {
-    if (!isWin) return null
+    if (!isSupportedOS) return null
 
     if (!SelectionService.instance) {
       SelectionService.instance = new SelectionService()
@@ -211,6 +216,8 @@ export class SelectionService {
       blacklist: SelectionHook!.FilterMode.EXCLUDE_LIST
     }
 
+    const predefinedBlacklist = isWin ? SELECTION_PREDEFINED_BLACKLIST.WINDOWS : SELECTION_PREDEFINED_BLACKLIST.MAC
+
     let combinedList: string[] = list
     let combinedMode = mode
 
@@ -219,7 +226,7 @@ export class SelectionService {
       switch (mode) {
         case 'blacklist':
           //combine the predefined blacklist with the user-defined blacklist
-          combinedList = [...new Set([...list, ...SELECTION_PREDEFINED_BLACKLIST.WINDOWS])]
+          combinedList = [...new Set([...list, ...predefinedBlacklist])]
           break
         case 'whitelist':
           combinedList = [...list]
@@ -227,7 +234,7 @@ export class SelectionService {
         case 'default':
         default:
           //use the predefined blacklist as the default filter list
-          combinedList = [...SELECTION_PREDEFINED_BLACKLIST.WINDOWS]
+          combinedList = [...predefinedBlacklist]
           combinedMode = 'blacklist'
           break
       }
@@ -241,14 +248,21 @@ export class SelectionService {
   private setHookFineTunedList() {
     if (!this.selectionHook) return
 
+    const excludeClipboardCursorDetectList = isWin
+      ? SELECTION_FINETUNED_LIST.EXCLUDE_CLIPBOARD_CURSOR_DETECT.WINDOWS
+      : SELECTION_FINETUNED_LIST.EXCLUDE_CLIPBOARD_CURSOR_DETECT.MAC
+    const includeClipboardDelayReadList = isWin
+      ? SELECTION_FINETUNED_LIST.INCLUDE_CLIPBOARD_DELAY_READ.WINDOWS
+      : SELECTION_FINETUNED_LIST.INCLUDE_CLIPBOARD_DELAY_READ.MAC
+
     this.selectionHook.setFineTunedList(
       SelectionHook!.FineTunedListType.EXCLUDE_CLIPBOARD_CURSOR_DETECT,
-      SELECTION_FINETUNED_LIST.EXCLUDE_CLIPBOARD_CURSOR_DETECT.WINDOWS
+      excludeClipboardCursorDetectList
     )
 
     this.selectionHook.setFineTunedList(
       SelectionHook!.FineTunedListType.INCLUDE_CLIPBOARD_DELAY_READ,
-      SELECTION_FINETUNED_LIST.INCLUDE_CLIPBOARD_DELAY_READ.WINDOWS
+      includeClipboardDelayReadList
     )
   }
 
@@ -257,11 +271,28 @@ export class SelectionService {
    * @returns {boolean} Success status of service start
    */
   public start(): boolean {
-    if (!this.selectionHook || this.started) {
-      this.logError(new Error('SelectionService start(): instance is null or already started'))
+    if (!this.selectionHook) {
+      this.logError(new Error('SelectionService start(): instance is null'))
       return false
     }
 
+    if (this.started) {
+      this.logError(new Error('SelectionService start(): already started'))
+      return false
+    }
+
+    //On macOS, we need to check if the process is trusted
+    if (isMac) {
+      if (!systemPreferences.isTrustedAccessibilityClient(false)) {
+        this.logError(
+          new Error(
+            'SelectionSerice not started: process is not trusted on macOS, please turn on the Accessibility permission'
+          )
+        )
+        return false
+      }
+    }
+
     try {
       //make sure the toolbar window is ready
       this.createToolbarWindow()
@@ -283,7 +314,7 @@ export class SelectionService {
         this.processTriggerMode()
 
         this.started = true
-        this.logInfo('SelectionService Started')
+        this.logInfo('SelectionService Started', true)
         return true
       }
 
@@ -304,6 +335,7 @@ export class SelectionService {
     if (!this.selectionHook) return false
 
     this.selectionHook.stop()
+
     this.selectionHook.cleanup() //already remove all listeners
 
     //reset the listener states
@@ -314,8 +346,11 @@ export class SelectionService {
       this.toolbarWindow.close()
       this.toolbarWindow = null
     }
+
+    this.closePreloadedActionWindows()
+
     this.started = false
-    this.logInfo('SelectionService Stopped')
+    this.logInfo('SelectionService Stopped', true)
     return true
   }
 
@@ -331,7 +366,22 @@ export class SelectionService {
     this.selectionHook = null
     this.initStatus = false
     SelectionService.instance = null
-    this.logInfo('SelectionService Quitted')
+    this.logInfo('SelectionService Quitted', true)
+  }
+
+  /**
+   * Toggle the enabled state of the selection service
+   * Will sync the new enabled store to all renderer windows
+   */
+  public toggleEnabled(enabled: boolean | undefined = undefined) {
+    if (!this.selectionHook) return
+
+    const newEnabled = enabled === undefined ? !configManager.getSelectionAssistantEnabled() : enabled
+
+    configManager.setSelectionAssistantEnabled(newEnabled)
+
+    //sync the new enabled state to all renderer windows
+    storeSyncService.syncToRenderer('selectionStore/setSelectionEnabled', newEnabled)
   }
 
   /**
@@ -347,21 +397,29 @@ export class SelectionService {
     this.toolbarWindow = new BrowserWindow({
       width: toolbarWidth,
       height: toolbarHeight,
+      show: false,
       frame: false,
       transparent: true,
       alwaysOnTop: true,
       skipTaskbar: true,
+      autoHideMenuBar: true,
       resizable: false,
       minimizable: false,
       maximizable: false,
+      fullscreenable: false, // [macOS] must be false
       movable: true,
-      focusable: false,
       hasShadow: false,
       thickFrame: false,
       roundedCorners: true,
       backgroundMaterial: 'none',
-      type: 'toolbar',
-      show: false,
+
+      // Platform specific settings
+      //   [macOS] DO NOT set type to 'panel', it will not work because it conflicts with other settings
+      //   [macOS] DO NOT set focusable to false, it will make other windows bring to front together
+      ...(isWin ? { type: 'toolbar', focusable: false } : {}),
+      hiddenInMissionControl: true, // [macOS only]
+      acceptFirstMouse: true, // [macOS only]
+
       webPreferences: {
         preload: join(__dirname, '../preload/index.js'),
         contextIsolation: true,
@@ -373,17 +431,29 @@ export class SelectionService {
 
     // Hide when losing focus
     this.toolbarWindow.on('blur', () => {
-      this.hideToolbar()
+      if (this.toolbarWindow!.isVisible()) {
+        this.hideToolbar()
+      }
     })
 
     // Clean up when closed
     this.toolbarWindow.on('closed', () => {
+      if (!this.toolbarWindow?.isDestroyed()) {
+        this.toolbarWindow?.destroy()
+      }
       this.toolbarWindow = null
     })
 
     // Add show/hide event listeners
     this.toolbarWindow.on('show', () => {
       this.toolbarWindow?.webContents.send(IpcChannel.Selection_ToolbarVisibilityChange, true)
+
+      // [macOS] force the toolbar window to be visible on current desktop
+      // but it will make docker icon flash. And we found that it's not necessary now.
+      // will remove after testing
+      // if (isMac) {
+      //   this.toolbarWindow!.setVisibleOnAllWorkspaces(false)
+      // }
     })
 
     this.toolbarWindow.on('hide', () => {
@@ -434,8 +504,29 @@ export class SelectionService {
       x: posX,
       y: posY
     })
-    this.toolbarWindow!.show()
-    this.toolbarWindow!.setOpacity(1)
+
+    //set the window to always on top (highest level)
+    //should set every time the window is shown
+    this.toolbarWindow!.setAlwaysOnTop(true, 'screen-saver')
+
+    // [macOS] force the toolbar window to be visible on current desktop
+    // but it will make docker icon flash. And we found that it's not necessary now.
+    // will remove after testing
+    // if (isMac) {
+    //   this.toolbarWindow!.setVisibleOnAllWorkspaces(true, { visibleOnFullScreen: true })
+    // }
+
+    // [macOS] MUST use `showInactive()` to prevent other windows bring to front together
+    // [Windows] is OK for both `show()` and `showInactive()` because of `focusable: false`
+    this.toolbarWindow!.showInactive()
+
+    /**
+     * [Windows]
+     *   In Windows 10, setOpacity(1) will make the window completely transparent
+     *   It's a strange behavior, so we don't use it for compatibility
+     */
+    // this.toolbarWindow!.setOpacity(1)
+
     this.startHideByMouseKeyListener()
   }
 
@@ -445,10 +536,52 @@ export class SelectionService {
   public hideToolbar(): void {
     if (!this.isToolbarAlive()) return
 
-    this.toolbarWindow!.setOpacity(0)
+    this.stopHideByMouseKeyListener()
+
+    // [Windows] just hide the toolbar window is enough
+    if (!isMac) {
+      this.toolbarWindow!.hide()
+      return
+    }
+
+    /************************************************
+     * [macOS] the following code is only for macOS
+     *************************************************/
+
+    // [macOS] a HACKY way
+    // make sure other windows do not bring to front when toolbar is hidden
+    // get all focusable windows and set them to not focusable
+    const focusableWindows: BrowserWindow[] = []
+    for (const window of BrowserWindow.getAllWindows()) {
+      if (!window.isDestroyed() && window.isVisible()) {
+        if (window.isFocusable()) {
+          focusableWindows.push(window)
+          window.setFocusable(false)
+        }
+      }
+    }
+
     this.toolbarWindow!.hide()
 
-    this.stopHideByMouseKeyListener()
+    // set them back to focusable after 50ms
+    setTimeout(() => {
+      for (const window of focusableWindows) {
+        if (!window.isDestroyed()) {
+          window.setFocusable(true)
+        }
+      }
+    }, 50)
+
+    // [macOS] hacky way
+    // Because toolbar is not a FOCUSED window, so the hover status will remain when next time show
+    // so we just send mouseMove event to the toolbar window to make the hover status disappear
+    this.toolbarWindow!.webContents.sendInputEvent({
+      type: 'mouseMove',
+      x: -1,
+      y: -1
+    })
+
+    return
   }
 
   /**
@@ -488,71 +621,71 @@ export class SelectionService {
   /**
    * Calculate optimal toolbar position based on selection context
    * Ensures toolbar stays within screen boundaries and follows selection direction
-   * @param point Reference point for positioning, must be INTEGER
+   * @param refPoint Reference point for positioning, must be INTEGER
    * @param orientation Preferred position relative to reference point
    * @returns Calculated screen coordinates for toolbar, INTEGER
    */
-  private calculateToolbarPosition(point: Point, orientation: RelativeOrientation): Point {
+  private calculateToolbarPosition(refPoint: Point, orientation: RelativeOrientation): Point {
     // Calculate initial position based on the specified anchor
-    let posX: number, posY: number
+    const posPoint: Point = { x: 0, y: 0 }
 
     const { toolbarWidth, toolbarHeight } = this.getToolbarRealSize()
 
     switch (orientation) {
       case 'topLeft':
-        posX = point.x - toolbarWidth
-        posY = point.y - toolbarHeight
+        posPoint.x = refPoint.x - toolbarWidth
+        posPoint.y = refPoint.y - toolbarHeight
         break
       case 'topRight':
-        posX = point.x
-        posY = point.y - toolbarHeight
+        posPoint.x = refPoint.x
+        posPoint.y = refPoint.y - toolbarHeight
         break
       case 'topMiddle':
-        posX = point.x - toolbarWidth / 2
-        posY = point.y - toolbarHeight
+        posPoint.x = refPoint.x - toolbarWidth / 2
+        posPoint.y = refPoint.y - toolbarHeight
         break
       case 'bottomLeft':
-        posX = point.x - toolbarWidth
-        posY = point.y
+        posPoint.x = refPoint.x - toolbarWidth
+        posPoint.y = refPoint.y
         break
       case 'bottomRight':
-        posX = point.x
-        posY = point.y
+        posPoint.x = refPoint.x
+        posPoint.y = refPoint.y
         break
       case 'bottomMiddle':
-        posX = point.x - toolbarWidth / 2
-        posY = point.y
+        posPoint.x = refPoint.x - toolbarWidth / 2
+        posPoint.y = refPoint.y
         break
       case 'middleLeft':
-        posX = point.x - toolbarWidth
-        posY = point.y - toolbarHeight / 2
+        posPoint.x = refPoint.x - toolbarWidth
+        posPoint.y = refPoint.y - toolbarHeight / 2
         break
       case 'middleRight':
-        posX = point.x
-        posY = point.y - toolbarHeight / 2
+        posPoint.x = refPoint.x
+        posPoint.y = refPoint.y - toolbarHeight / 2
         break
       case 'center':
-        posX = point.x - toolbarWidth / 2
-        posY = point.y - toolbarHeight / 2
+        posPoint.x = refPoint.x - toolbarWidth / 2
+        posPoint.y = refPoint.y - toolbarHeight / 2
         break
       default:
         // Default to 'topMiddle' if invalid position
-        posX = point.x - toolbarWidth / 2
-        posY = point.y - toolbarHeight / 2
+        posPoint.x = refPoint.x - toolbarWidth / 2
+        posPoint.y = refPoint.y - toolbarHeight / 2
     }
 
     //use original point to get the display
-    const display = screen.getDisplayNearestPoint({ x: point.x, y: point.y })
+    const display = screen.getDisplayNearestPoint(refPoint)
 
     // Ensure toolbar stays within screen boundaries
-    posX = Math.round(
-      Math.max(display.workArea.x, Math.min(posX, display.workArea.x + display.workArea.width - toolbarWidth))
+    posPoint.x = Math.round(
+      Math.max(display.workArea.x, Math.min(posPoint.x, display.workArea.x + display.workArea.width - toolbarWidth))
     )
-    posY = Math.round(
-      Math.max(display.workArea.y, Math.min(posY, display.workArea.y + display.workArea.height - toolbarHeight))
+    posPoint.y = Math.round(
+      Math.max(display.workArea.y, Math.min(posPoint.y, display.workArea.y + display.workArea.height - toolbarHeight))
     )
 
-    return { x: posX, y: posY }
+    return posPoint
   }
 
   private isSamePoint(point1: Point, point2: Point): boolean {
@@ -563,6 +696,21 @@ export class SelectionService {
     return startTop.y === endTop.y && startBottom.y === endBottom.y
   }
 
+  /**
+   * Get the user selected text and process it (trigger by shortcut)
+   *
+   * it's a public method used by shortcut service
+   */
+  public processSelectTextByShortcut(): void {
+    if (!this.selectionHook || !this.started || this.triggerMode !== TriggerMode.Shortcut) return
+
+    const selectionData = this.selectionHook.getCurrentSelection()
+
+    if (selectionData) {
+      this.processTextSelection(selectionData)
+    }
+  }
+
   /**
    * Determine if the text selection should be processed by filter mode&list
    * @param selectionData Text selection information and coordinates
@@ -726,8 +874,11 @@ export class SelectionService {
     }
 
     if (!isLogical) {
+      // [macOS] don't need to convert by screenToDipPoint
+      if (!isMac) {
+        refPoint = screen.screenToDipPoint(refPoint)
+      }
       //screenToDipPoint can be float, so we need to round it
-      refPoint = screen.screenToDipPoint(refPoint)
       refPoint = { x: Math.round(refPoint.x), y: Math.round(refPoint.y) }
     }
 
@@ -785,8 +936,8 @@ export class SelectionService {
       return
     }
 
-    //data point is physical coordinates, convert to logical coordinates
-    const mousePoint = screen.screenToDipPoint({ x: data.x, y: data.y })
+    //data point is physical coordinates, convert to logical coordinates(only for windows/linux)
+    const mousePoint = isMac ? { x: data.x, y: data.y } : screen.screenToDipPoint({ x: data.x, y: data.y })
 
     const bounds = this.toolbarWindow!.getBounds()
 
@@ -854,7 +1005,6 @@ export class SelectionService {
     this.lastCtrlkeyDownTime = -1
 
     const selectionData = this.selectionHook!.getCurrentSelection()
-
     if (selectionData) {
       this.processTextSelection(selectionData)
     }
@@ -920,7 +1070,8 @@ export class SelectionService {
       frame: false,
       transparent: true,
       autoHideMenuBar: true,
-      titleBarStyle: 'hidden',
+      titleBarStyle: 'hidden', // [macOS]
+      trafficLightPosition: { x: 12, y: 9 }, // [macOS]
       hasShadow: false,
       thickFrame: false,
       show: false,
@@ -958,6 +1109,17 @@ export class SelectionService {
     }
   }
 
+  /**
+   * Close all preloaded action windows
+   */
+  private closePreloadedActionWindows() {
+    for (const actionWindow of this.preloadedActionWindows) {
+      if (!actionWindow.isDestroyed()) {
+        actionWindow.destroy()
+      }
+    }
+  }
+
   /**
    * Preload a new action window asynchronously
    * This method is called after popping a window to ensure we always have windows ready
@@ -986,6 +1148,27 @@ export class SelectionService {
       if (!actionWindow.isDestroyed()) {
         actionWindow.destroy()
       }
+
+      // [macOS] a HACKY way
+      // make sure other windows do not bring to front when action window is closed
+      if (isMac) {
+        const focusableWindows: BrowserWindow[] = []
+        for (const window of BrowserWindow.getAllWindows()) {
+          if (!window.isDestroyed() && window.isVisible()) {
+            if (window.isFocusable()) {
+              focusableWindows.push(window)
+              window.setFocusable(false)
+            }
+          }
+        }
+        setTimeout(() => {
+          for (const window of focusableWindows) {
+            if (!window.isDestroyed()) {
+              window.setFocusable(true)
+            }
+          }
+        }, 50)
+      }
     })
 
     //remember the action window size
@@ -1031,22 +1214,26 @@ export class SelectionService {
 
     //center way
     if (!this.isFollowToolbar || !this.toolbarWindow) {
-      if (this.isRemeberWinSize) {
-        actionWindow.setBounds({
-          width: actionWindowWidth,
-          height: actionWindowHeight
-        })
-      }
+      const display = screen.getDisplayNearestPoint(screen.getCursorScreenPoint())
+      const workArea = display.workArea
+
+      const centerX = workArea.x + (workArea.width - actionWindowWidth) / 2
+      const centerY = workArea.y + (workArea.height - actionWindowHeight) / 2
+
+      actionWindow.setBounds({
+        width: actionWindowWidth,
+        height: actionWindowHeight,
+        x: Math.round(centerX),
+        y: Math.round(centerY)
+      })
 
       actionWindow.show()
-      this.hideToolbar()
       return
     }
 
     //follow toolbar
-
     const toolbarBounds = this.toolbarWindow!.getBounds()
-    const display = screen.getDisplayNearestPoint({ x: toolbarBounds.x, y: toolbarBounds.y })
+    const display = screen.getDisplayNearestPoint(screen.getCursorScreenPoint())
     const workArea = display.workArea
     const GAP = 6 // 6px gap from screen edges
 
@@ -1106,29 +1293,44 @@ export class SelectionService {
    * Manages appropriate event listeners for each mode
    */
   private processTriggerMode() {
-    if (this.triggerMode === TriggerMode.Selected) {
-      if (this.isCtrlkeyListenerActive) {
-        this.selectionHook!.off('key-down', this.handleKeyDownCtrlkeyMode)
-        this.selectionHook!.off('key-up', this.handleKeyUpCtrlkeyMode)
+    switch (this.triggerMode) {
+      case TriggerMode.Selected:
+        if (this.isCtrlkeyListenerActive) {
+          this.selectionHook!.off('key-down', this.handleKeyDownCtrlkeyMode)
+          this.selectionHook!.off('key-up', this.handleKeyUpCtrlkeyMode)
 
-        this.isCtrlkeyListenerActive = false
-      }
+          this.isCtrlkeyListenerActive = false
+        }
 
-      this.selectionHook!.setSelectionPassiveMode(false)
-    } else if (this.triggerMode === TriggerMode.Ctrlkey) {
-      if (!this.isCtrlkeyListenerActive) {
-        this.selectionHook!.on('key-down', this.handleKeyDownCtrlkeyMode)
-        this.selectionHook!.on('key-up', this.handleKeyUpCtrlkeyMode)
+        this.selectionHook!.setSelectionPassiveMode(false)
+        break
+      case TriggerMode.Ctrlkey:
+        if (!this.isCtrlkeyListenerActive) {
+          this.selectionHook!.on('key-down', this.handleKeyDownCtrlkeyMode)
+          this.selectionHook!.on('key-up', this.handleKeyUpCtrlkeyMode)
 
-        this.isCtrlkeyListenerActive = true
-      }
+          this.isCtrlkeyListenerActive = true
+        }
 
-      this.selectionHook!.setSelectionPassiveMode(true)
+        this.selectionHook!.setSelectionPassiveMode(true)
+        break
+      case TriggerMode.Shortcut:
+        //remove the ctrlkey listener, don't need any key listener for shortcut mode
+        if (this.isCtrlkeyListenerActive) {
+          this.selectionHook!.off('key-down', this.handleKeyDownCtrlkeyMode)
+          this.selectionHook!.off('key-up', this.handleKeyUpCtrlkeyMode)
+
+          this.isCtrlkeyListenerActive = false
+        }
+
+        this.selectionHook!.setSelectionPassiveMode(true)
+        break
     }
   }
 
   public writeToClipboard(text: string): boolean {
-    return this.selectionHook?.writeToClipboard(text) ?? false
+    if (!this.selectionHook || !this.started) return false
+    return this.selectionHook.writeToClipboard(text)
   }
 
   /**
@@ -1142,7 +1344,7 @@ export class SelectionService {
       selectionService?.hideToolbar()
     })
 
-    ipcMain.handle(IpcChannel.Selection_WriteToClipboard, (_, text: string) => {
+    ipcMain.handle(IpcChannel.Selection_WriteToClipboard, (_, text: string): boolean => {
       return selectionService?.writeToClipboard(text) ?? false
     })
 
@@ -1202,8 +1404,10 @@ export class SelectionService {
     this.isIpcHandlerRegistered = true
   }
 
-  private logInfo(message: string) {
-    isDev && Logger.info('[SelectionService] Info: ', message)
+  private logInfo(message: string, forceShow: boolean = false) {
+    if (isDev || forceShow) {
+      Logger.info('[SelectionService] Info: ', message)
+    }
   }
 
   private logError(...args: [...string[], Error]) {
@@ -1217,7 +1421,7 @@ export class SelectionService {
  * @returns {boolean} Success status of initialization
  */
 export function initSelectionService(): boolean {
-  if (!isWin) return false
+  if (!isSupportedOS) return false
 
   configManager.subscribe(ConfigKeys.SelectionAssistantEnabled, (enabled: boolean) => {
     //avoid closure

src/main/services/ShortcutService.ts

@@ -4,10 +4,16 @@ import { BrowserWindow, globalShortcut } from 'electron'
 import Logger from 'electron-log'
 
 import { configManager } from './ConfigManager'
+import selectionService from './SelectionService'
 import { windowService } from './WindowService'
 
 let showAppAccelerator: string | null = null
 let showMiniWindowAccelerator: string | null = null
+let selectionAssistantToggleAccelerator: string | null = null
+let selectionAssistantSelectTextAccelerator: string | null = null
+
+//indicate if the shortcuts are registered on app boot time
+let isRegisterOnBoot = true
 
 // store the focus and blur handlers for each window to unregister them later
 const windowOnHandlers = new Map<BrowserWindow, { onFocusHandler: () => void; onBlurHandler: () => void }>()
@@ -28,6 +34,18 @@ function getShortcutHandler(shortcut: Shortcut) {
       return () => {
         windowService.toggleMiniWindow()
       }
+    case 'selection_assistant_toggle':
+      return () => {
+        if (selectionService) {
+          selectionService.toggleEnabled()
+        }
+      }
+    case 'selection_assistant_select_text':
+      return () => {
+        if (selectionService) {
+          selectionService.processSelectTextByShortcut()
+        }
+      }
     default:
       return null
   }
@@ -37,9 +55,8 @@ function formatShortcutKey(shortcut: string[]): string {
   return shortcut.join('+')
 }
 
-const convertShortcutRecordedByKeyboardEventKeyValueToElectronGlobalShortcutFormat = (
-  shortcut: string | string[]
-): string => {
+// convert the shortcut recorded by keyboard event key value to electron global shortcut format
+const convertShortcutFormat = (shortcut: string | string[]): string => {
   const accelerator = (() => {
     if (Array.isArray(shortcut)) {
       return shortcut
@@ -93,11 +110,14 @@ const convertShortcutRecordedByKeyboardEventKeyValueToElectronGlobalShortcutForm
 }
 
 export function registerShortcuts(window: BrowserWindow) {
-  window.once('ready-to-show', () => {
-    if (configManager.getLaunchToTray()) {
-      registerOnlyUniversalShortcuts()
-    }
-  })
+  if (isRegisterOnBoot) {
+    window.once('ready-to-show', () => {
+      if (configManager.getLaunchToTray()) {
+        registerOnlyUniversalShortcuts()
+      }
+    })
+    isRegisterOnBoot = false
+  }
 
   //only for clearer code
   const registerOnlyUniversalShortcuts = () => {
@@ -124,7 +144,12 @@ export function registerShortcuts(window: BrowserWindow) {
         }
 
         // only register universal shortcuts when needed
-        if (onlyUniversalShortcuts && !['show_app', 'mini_window'].includes(shortcut.key)) {
+        if (
+          onlyUniversalShortcuts &&
+          !['show_app', 'mini_window', 'selection_assistant_toggle', 'selection_assistant_select_text'].includes(
+            shortcut.key
+          )
+        ) {
           return
         }
 
@@ -146,6 +171,14 @@ export function registerShortcuts(window: BrowserWindow) {
             showMiniWindowAccelerator = formatShortcutKey(shortcut.shortcut)
             break
 
+          case 'selection_assistant_toggle':
+            selectionAssistantToggleAccelerator = formatShortcutKey(shortcut.shortcut)
+            break
+
+          case 'selection_assistant_select_text':
+            selectionAssistantSelectTextAccelerator = formatShortcutKey(shortcut.shortcut)
+            break
+
           //the following ZOOMs will register shortcuts seperately, so will return
           case 'zoom_in':
             globalShortcut.register('CommandOrControl+=', () => handler(window))
@@ -162,9 +195,7 @@ export function registerShortcuts(window: BrowserWindow) {
             return
         }
 
-        const accelerator = convertShortcutRecordedByKeyboardEventKeyValueToElectronGlobalShortcutFormat(
-          shortcut.shortcut
-        )
+        const accelerator = convertShortcutFormat(shortcut.shortcut)
 
         globalShortcut.register(accelerator, () => handler(window))
       } catch (error) {
@@ -181,15 +212,25 @@ export function registerShortcuts(window: BrowserWindow) {
 
       if (showAppAccelerator) {
         const handler = getShortcutHandler({ key: 'show_app' } as Shortcut)
-        const accelerator =
-          convertShortcutRecordedByKeyboardEventKeyValueToElectronGlobalShortcutFormat(showAppAccelerator)
+        const accelerator = convertShortcutFormat(showAppAccelerator)
         handler && globalShortcut.register(accelerator, () => handler(window))
       }
 
       if (showMiniWindowAccelerator) {
         const handler = getShortcutHandler({ key: 'mini_window' } as Shortcut)
-        const accelerator =
-          convertShortcutRecordedByKeyboardEventKeyValueToElectronGlobalShortcutFormat(showMiniWindowAccelerator)
+        const accelerator = convertShortcutFormat(showMiniWindowAccelerator)
+        handler && globalShortcut.register(accelerator, () => handler(window))
+      }
+
+      if (selectionAssistantToggleAccelerator) {
+        const handler = getShortcutHandler({ key: 'selection_assistant_toggle' } as Shortcut)
+        const accelerator = convertShortcutFormat(selectionAssistantToggleAccelerator)
+        handler && globalShortcut.register(accelerator, () => handler(window))
+      }
+
+      if (selectionAssistantSelectTextAccelerator) {
+        const handler = getShortcutHandler({ key: 'selection_assistant_select_text' } as Shortcut)
+        const accelerator = convertShortcutFormat(selectionAssistantSelectTextAccelerator)
         handler && globalShortcut.register(accelerator, () => handler(window))
       }
     } catch (error) {
@@ -217,6 +258,8 @@ export function unregisterAllShortcuts() {
   try {
     showAppAccelerator = null
     showMiniWindowAccelerator = null
+    selectionAssistantToggleAccelerator = null
+    selectionAssistantSelectTextAccelerator = null
     windowOnHandlers.forEach((handlers, window) => {
       window.off('focus', handlers.onFocusHandler)
       window.off('blur', handlers.onBlurHandler)

src/main/services/StoreSyncService.ts

@@ -49,6 +49,23 @@ export class StoreSyncService {
     this.windowIds = this.windowIds.filter((id) => id !== windowId)
   }
 
+  /**
+   * Sync an action to all renderer windows
+   * @param type Action type, like 'settings/setTray'
+   * @param payload Action payload
+   *
+   * NOTICE: DO NOT use directly in ConfigManager, may cause infinite sync loop
+   */
+  public syncToRenderer(type: string, payload: any): void {
+    const action: StoreSyncAction = {
+      type,
+      payload
+    }
+
+    //-1 means the action is from the main process, will be broadcast to all windows
+    this.broadcastToOtherWindows(-1, action)
+  }
+
   /**
    * Register IPC handlers for store sync communication
    * Handles window subscription, unsubscription and action broadcasting

src/main/services/TrayService.ts

@@ -1,4 +1,4 @@
-import { isMac } from '@main/constant'
+import { isLinux, isMac, isWin } from '@main/constant'
 import { locales } from '@main/utils/locales'
 import { app, Menu, MenuItemConstructorOptions, nativeImage, nativeTheme, Tray } from 'electron'
 
@@ -6,6 +6,7 @@ import icon from '../../../build/tray_icon.png?asset'
 import iconDark from '../../../build/tray_icon_dark.png?asset'
 import iconLight from '../../../build/tray_icon_light.png?asset'
 import { ConfigKeys, configManager } from './ConfigManager'
+import selectionService from './SelectionService'
 import { windowService } from './WindowService'
 
 export class TrayService {
@@ -29,14 +30,14 @@ export class TrayService {
     const iconPath = isMac ? (nativeTheme.shouldUseDarkColors ? iconLight : iconDark) : icon
     const tray = new Tray(iconPath)
 
-    if (process.platform === 'win32') {
+    if (isWin) {
       tray.setImage(iconPath)
-    } else if (process.platform === 'darwin') {
+    } else if (isMac) {
       const image = nativeImage.createFromPath(iconPath)
       const resizedImage = image.resize({ width: 16, height: 16 })
       resizedImage.setTemplateImage(true)
       tray.setImage(resizedImage)
-    } else if (process.platform === 'linux') {
+    } else if (isLinux) {
       const image = nativeImage.createFromPath(iconPath)
       const resizedImage = image.resize({ width: 16, height: 16 })
       tray.setImage(resizedImage)
@@ -46,7 +47,7 @@ export class TrayService {
 
     this.updateContextMenu()
 
-    if (process.platform === 'linux') {
+    if (isLinux) {
       this.tray.setContextMenu(this.contextMenu)
     }
 
@@ -69,19 +70,29 @@ export class TrayService {
 
   private updateContextMenu() {
     const locale = locales[configManager.getLanguage()]
-    const { tray: trayLocale } = locale.translation
+    const { tray: trayLocale, selection: selectionLocale } = locale.translation
 
-    const enableQuickAssistant = configManager.getEnableQuickAssistant()
+    const quickAssistantEnabled = configManager.getEnableQuickAssistant()
+    const selectionAssistantEnabled = configManager.getSelectionAssistantEnabled()
 
     const template = [
       {
         label: trayLocale.show_window,
         click: () => windowService.showMainWindow()
       },
-      enableQuickAssistant && {
+      quickAssistantEnabled && {
         label: trayLocale.show_mini_window,
         click: () => windowService.showMiniWindow()
       },
+      (isWin || isMac) && {
+        label: selectionLocale.name + (selectionAssistantEnabled ? ' - On' : ' - Off'),
+        click: () => {
+          if (selectionService) {
+            selectionService.toggleEnabled()
+            this.updateContextMenu()
+          }
+        }
+      },
       { type: 'separator' },
       {
         label: trayLocale.quit,
@@ -118,6 +129,10 @@ export class TrayService {
     configManager.subscribe(ConfigKeys.EnableQuickAssistant, () => {
       this.updateContextMenu()
     })
+
+    configManager.subscribe(ConfigKeys.SelectionAssistantEnabled, () => {
+      this.updateContextMenu()
+    })
   }
 
   private quit() {

src/main/services/VertexAIService.ts

@@ -0,0 +1,142 @@
+import { GoogleAuth } from 'google-auth-library'
+
+interface ServiceAccountCredentials {
+  privateKey: string
+  clientEmail: string
+}
+
+interface VertexAIAuthParams {
+  projectId: string
+  serviceAccount?: ServiceAccountCredentials
+}
+
+const REQUIRED_VERTEX_AI_SCOPE = 'https://www.googleapis.com/auth/cloud-platform'
+
+class VertexAIService {
+  private static instance: VertexAIService
+  private authClients: Map<string, GoogleAuth> = new Map()
+
+  static getInstance(): VertexAIService {
+    if (!VertexAIService.instance) {
+      VertexAIService.instance = new VertexAIService()
+    }
+    return VertexAIService.instance
+  }
+
+  /**
+   * 格式化私钥，确保它包含正确的PEM头部和尾部
+   */
+  private formatPrivateKey(privateKey: string): string {
+    if (!privateKey || typeof privateKey !== 'string') {
+      throw new Error('Private key must be a non-empty string')
+    }
+
+    // 处理JSON字符串中的转义换行符
+    let key = privateKey.replace(/\\n/g, '\n')
+
+    // 如果已经是正确格式的PEM，直接返回
+    if (key.includes('-----BEGIN PRIVATE KEY-----') && key.includes('-----END PRIVATE KEY-----')) {
+      return key
+    }
+
+    // 移除所有换行符和空白字符（为了重新格式化）
+    key = key.replace(/\s+/g, '')
+
+    // 移除可能存在的头部和尾部
+    key = key.replace(/-----BEGIN[^-]*-----/g, '')
+    key = key.replace(/-----END[^-]*-----/g, '')
+
+    // 确保私钥不为空
+    if (!key) {
+      throw new Error('Private key is empty after formatting')
+    }
+
+    // 添加正确的PEM头部和尾部，并格式化为64字符一行
+    const formattedKey = key.match(/.{1,64}/g)?.join('\n') || key
+
+    return `-----BEGIN PRIVATE KEY-----\n${formattedKey}\n-----END PRIVATE KEY-----`
+  }
+
+  /**
+   * 获取认证头用于 Vertex AI 请求
+   */
+  async getAuthHeaders(params: VertexAIAuthParams): Promise<Record<string, string>> {
+    const { projectId, serviceAccount } = params
+
+    if (!serviceAccount?.privateKey || !serviceAccount?.clientEmail) {
+      throw new Error('Service account credentials are required')
+    }
+
+    // 创建缓存键
+    const cacheKey = `${projectId}-${serviceAccount.clientEmail}`
+
+    // 检查是否已有客户端实例
+    let auth = this.authClients.get(cacheKey)
+
+    if (!auth) {
+      try {
+        // 格式化私钥
+        const formattedPrivateKey = this.formatPrivateKey(serviceAccount.privateKey)
+
+        // 创建新的认证客户端
+        auth = new GoogleAuth({
+          credentials: {
+            private_key: formattedPrivateKey,
+            client_email: serviceAccount.clientEmail
+          },
+          projectId,
+          scopes: [REQUIRED_VERTEX_AI_SCOPE]
+        })
+
+        this.authClients.set(cacheKey, auth)
+      } catch (formatError: any) {
+        throw new Error(`Invalid private key format: ${formatError.message}`)
+      }
+    }
+
+    try {
+      // 获取认证头
+      const authHeaders = await auth.getRequestHeaders()
+
+      // 转换为普通对象
+      const headers: Record<string, string> = {}
+      for (const [key, value] of Object.entries(authHeaders)) {
+        if (typeof value === 'string') {
+          headers[key] = value
+        }
+      }
+
+      return headers
+    } catch (error: any) {
+      // 如果认证失败，清除缓存的客户端
+      this.authClients.delete(cacheKey)
+      throw new Error(`Failed to authenticate with service account: ${error.message}`)
+    }
+  }
+
+  /**
+   * 清理指定项目的认证缓存
+   */
+  clearAuthCache(projectId: string, clientEmail?: string): void {
+    if (clientEmail) {
+      const cacheKey = `${projectId}-${clientEmail}`
+      this.authClients.delete(cacheKey)
+    } else {
+      // 清理该项目的所有缓存
+      for (const [key] of this.authClients) {
+        if (key.startsWith(`${projectId}-`)) {
+          this.authClients.delete(key)
+        }
+      }
+    }
+  }
+
+  /**
+   * 清理所有认证缓存
+   */
+  clearAllAuthCache(): void {
+    this.authClients.clear()
+  }
+}
+
+export default VertexAIService

src/main/services/WindowService.ts

@@ -63,7 +63,7 @@ export class WindowService {
       titleBarOverlay: nativeTheme.shouldUseDarkColors ? titleBarOverlayDark : titleBarOverlayLight,
       backgroundColor: isMac ? undefined : nativeTheme.shouldUseDarkColors ? '#181818' : '#FFFFFF',
       darkTheme: nativeTheme.shouldUseDarkColors,
-      trafficLightPosition: { x: 12, y: 12 },
+      trafficLightPosition: { x: 8, y: 12 },
       ...(isLinux ? { icon } : {}),
       webPreferences: {
         preload: join(__dirname, '../preload/index.js'),
@@ -95,6 +95,7 @@ export class WindowService {
 
     this.setupMaximize(mainWindow, mainWindowState.isMaximized)
     this.setupContextMenu(mainWindow)
+    this.setupSpellCheck(mainWindow)
     this.setupWindowEvents(mainWindow)
     this.setupWebContentsHandlers(mainWindow)
     this.setupWindowLifecycleEvents(mainWindow)
@@ -102,6 +103,18 @@ export class WindowService {
     this.loadMainWindowContent(mainWindow)
   }
 
+  private setupSpellCheck(mainWindow: BrowserWindow) {
+    const enableSpellCheck = configManager.get('enableSpellCheck', false)
+    if (enableSpellCheck) {
+      try {
+        const spellCheckLanguages = configManager.get('spellCheckLanguages', []) as string[]
+        spellCheckLanguages.length > 0 && mainWindow.webContents.session.setSpellCheckerLanguages(spellCheckLanguages)
+      } catch (error) {
+        Logger.error('Failed to set spell check languages:', error as Error)
+      }
+    }
+  }
+
   private setupMainWindowMonitor(mainWindow: BrowserWindow) {
     mainWindow.webContents.on('render-process-gone', (_, details) => {
       Logger.error(`Renderer process crashed with: ${JSON.stringify(details)}`)
@@ -116,12 +129,6 @@ export class WindowService {
         app.exit(1)
       }
     })
-
-    mainWindow.webContents.on('unresponsive', () => {
-      // 在升级到electron 34后，可以获取具体js stack trace,目前只打个日志监控下
-      // https://www.electronjs.org/blog/electron-34-0#unresponsive-renderer-javascript-call-stacks
-      Logger.error('Renderer process unresponsive')
-    })
   }
 
   private setupMaximize(mainWindow: BrowserWindow, isMaximized: boolean) {
@@ -136,9 +143,10 @@ export class WindowService {
   }
 
   private setupContextMenu(mainWindow: BrowserWindow) {
-    contextMenu.contextMenu(mainWindow)
-    app.on('browser-window-created', (_, win) => {
-      contextMenu.contextMenu(win)
+    contextMenu.contextMenu(mainWindow.webContents)
+    // setup context menu for all webviews like miniapp
+    app.on('web-contents-created', (_, webContents) => {
+      contextMenu.contextMenu(webContents)
     })
 
     // Dangerous API
@@ -443,8 +451,7 @@ export class WindowService {
         preload: join(__dirname, '../preload/index.js'),
         sandbox: false,
         webSecurity: false,
-        webviewTag: true,
-        backgroundThrottling: false
+        webviewTag: true
       }
     })
 

src/main/services/mcp/oauth/provider.ts

@@ -1,12 +1,8 @@
 import path from 'node:path'
 
 import { getConfigDir } from '@main/utils/file'
-import { OAuthClientProvider } from '@modelcontextprotocol/sdk/client/auth.js'
-import {
-  OAuthClientInformation,
-  OAuthClientInformationFull,
-  OAuthTokens
-} from '@modelcontextprotocol/sdk/shared/auth.js'
+import { OAuthClientProvider } from '@modelcontextprotocol/sdk/client/auth'
+import { OAuthClientInformation, OAuthClientInformationFull, OAuthTokens } from '@modelcontextprotocol/sdk/shared/auth'
 import Logger from 'electron-log'
 import open from 'open'
 

src/main/services/remotefile/BaseFileService.ts

@@ -0,0 +1,13 @@
+import { FileListResponse, FileMetadata, FileUploadResponse, Provider } from '@types'
+
+export abstract class BaseFileService {
+  protected readonly provider: Provider
+  protected constructor(provider: Provider) {
+    this.provider = provider
+  }
+
+  abstract uploadFile(file: FileMetadata): Promise<FileUploadResponse>
+  abstract deleteFile(fileId: string): Promise<void>
+  abstract listFiles(): Promise<FileListResponse>
+  abstract retrieveFile(fileId: string): Promise<FileUploadResponse>
+}

src/main/services/remotefile/FileServiceManager.ts

@@ -0,0 +1,41 @@
+import { Provider } from '@types'
+
+import { BaseFileService } from './BaseFileService'
+import { GeminiService } from './GeminiService'
+import { MistralService } from './MistralService'
+
+export class FileServiceManager {
+  private static instance: FileServiceManager
+  private services: Map<string, BaseFileService> = new Map()
+
+  // eslint-disable-next-line @typescript-eslint/no-empty-function
+  private constructor() {}
+
+  static getInstance(): FileServiceManager {
+    if (!this.instance) {
+      this.instance = new FileServiceManager()
+    }
+    return this.instance
+  }
+
+  getService(provider: Provider): BaseFileService {
+    const type = provider.type
+    let service = this.services.get(type)
+
+    if (!service) {
+      switch (type) {
+        case 'gemini':
+          service = new GeminiService(provider)
+          break
+        case 'mistral':
+          service = new MistralService(provider)
+          break
+        default:
+          throw new Error(`Unsupported service type: ${type}`)
+      }
+      this.services.set(type, service)
+    }
+
+    return service
+  }
+}

src/main/services/remotefile/GeminiService.ts

@@ -0,0 +1,190 @@
+import { File, Files, FileState, GoogleGenAI } from '@google/genai'
+import { FileListResponse, FileMetadata, FileUploadResponse, Provider } from '@types'
+import Logger from 'electron-log'
+import { v4 as uuidv4 } from 'uuid'
+
+import { CacheService } from '../CacheService'
+import { BaseFileService } from './BaseFileService'
+
+export class GeminiService extends BaseFileService {
+  private static readonly FILE_LIST_CACHE_KEY = 'gemini_file_list'
+  private static readonly FILE_CACHE_DURATION = 48 * 60 * 60 * 1000
+  private static readonly LIST_CACHE_DURATION = 3000
+
+  protected readonly fileManager: Files
+
+  constructor(provider: Provider) {
+    super(provider)
+    this.fileManager = new GoogleGenAI({
+      vertexai: false,
+      apiKey: provider.apiKey,
+      httpOptions: {
+        baseUrl: provider.apiHost
+      }
+    }).files
+  }
+
+  async uploadFile(file: FileMetadata): Promise<FileUploadResponse> {
+    try {
+      const uploadResult = await this.fileManager.upload({
+        file: file.path,
+        config: {
+          mimeType: 'application/pdf',
+          name: file.id,
+          displayName: file.origin_name
+        }
+      })
+
+      // 根据文件状态设置响应状态
+      let status: 'success' | 'processing' | 'failed' | 'unknown'
+      switch (uploadResult.state) {
+        case FileState.ACTIVE:
+          status = 'success'
+          break
+        case FileState.PROCESSING:
+          status = 'processing'
+          break
+        case FileState.FAILED:
+          status = 'failed'
+          break
+        default:
+          status = 'unknown'
+      }
+
+      const response: FileUploadResponse = {
+        fileId: uploadResult.name || '',
+        displayName: file.origin_name,
+        status,
+        originalFile: {
+          type: 'gemini',
+          file: uploadResult
+        }
+      }
+
+      // 只缓存成功的文件
+      if (status === 'success') {
+        const cacheKey = `${GeminiService.FILE_LIST_CACHE_KEY}_${response.fileId}`
+        CacheService.set<FileUploadResponse>(cacheKey, response, GeminiService.FILE_CACHE_DURATION)
+      }
+
+      return response
+    } catch (error) {
+      Logger.error('Error uploading file to Gemini:', error)
+      return {
+        fileId: '',
+        displayName: file.origin_name,
+        status: 'failed',
+        originalFile: undefined
+      }
+    }
+  }
+
+  async retrieveFile(fileId: string): Promise<FileUploadResponse> {
+    try {
+      const cachedResponse = CacheService.get<FileUploadResponse>(`${GeminiService.FILE_LIST_CACHE_KEY}_${fileId}`)
+      Logger.info('[GeminiService] cachedResponse', cachedResponse)
+      if (cachedResponse) {
+        return cachedResponse
+      }
+      const files: File[] = []
+
+      for await (const f of await this.fileManager.list()) {
+        files.push(f)
+      }
+      Logger.info('[GeminiService] files', files)
+      const file = files
+        .filter((file) => file.state === FileState.ACTIVE)
+        .find((file) => file.name?.substring(6) === fileId) // 去掉 files/ 前缀
+      Logger.info('[GeminiService] file', file)
+      if (file) {
+        return {
+          fileId: fileId,
+          displayName: file.displayName || '',
+          status: 'success',
+          originalFile: {
+            type: 'gemini',
+            file
+          }
+        }
+      }
+
+      return {
+        fileId: fileId,
+        displayName: '',
+        status: 'failed',
+        originalFile: undefined
+      }
+    } catch (error) {
+      Logger.error('Error retrieving file from Gemini:', error)
+      return {
+        fileId: fileId,
+        displayName: '',
+        status: 'failed',
+        originalFile: undefined
+      }
+    }
+  }
+
+  async listFiles(): Promise<FileListResponse> {
+    try {
+      const cachedList = CacheService.get<FileListResponse>(GeminiService.FILE_LIST_CACHE_KEY)
+      if (cachedList) {
+        return cachedList
+      }
+      const geminiFiles: File[] = []
+
+      for await (const f of await this.fileManager.list()) {
+        geminiFiles.push(f)
+      }
+      const fileList: FileListResponse = {
+        files: geminiFiles
+          .filter((file) => file.state === FileState.ACTIVE)
+          .map((file) => {
+            // 更新单个文件的缓存
+            const fileResponse: FileUploadResponse = {
+              fileId: file.name || uuidv4(),
+              displayName: file.displayName || '',
+              status: 'success',
+              originalFile: {
+                type: 'gemini',
+                file
+              }
+            }
+            CacheService.set(
+              `${GeminiService.FILE_LIST_CACHE_KEY}_${file.name}`,
+              fileResponse,
+              GeminiService.FILE_CACHE_DURATION
+            )
+
+            return {
+              id: file.name || uuidv4(),
+              displayName: file.displayName || '',
+              size: Number(file.sizeBytes),
+              status: 'success',
+              originalFile: {
+                type: 'gemini',
+                file
+              }
+            }
+          })
+      }
+
+      // 更新文件列表缓存
+      CacheService.set(GeminiService.FILE_LIST_CACHE_KEY, fileList, GeminiService.LIST_CACHE_DURATION)
+      return fileList
+    } catch (error) {
+      Logger.error('Error listing files from Gemini:', error)
+      return { files: [] }
+    }
+  }
+
+  async deleteFile(fileId: string): Promise<void> {
+    try {
+      await this.fileManager.delete({ name: fileId })
+      Logger.info(`File ${fileId} deleted from Gemini`)
+    } catch (error) {
+      Logger.error('Error deleting file from Gemini:', error)
+      throw error
+    }
+  }
+}

src/main/services/remotefile/MistralService.ts

@@ -0,0 +1,104 @@
+import fs from 'node:fs/promises'
+
+import { Mistral } from '@mistralai/mistralai'
+import { FileListResponse, FileMetadata, FileUploadResponse, Provider } from '@types'
+import Logger from 'electron-log'
+
+import { MistralClientManager } from '../MistralClientManager'
+import { BaseFileService } from './BaseFileService'
+
+export class MistralService extends BaseFileService {
+  private readonly client: Mistral
+
+  constructor(provider: Provider) {
+    super(provider)
+    const clientManager = MistralClientManager.getInstance()
+    clientManager.initializeClient(provider)
+    this.client = clientManager.getClient()
+  }
+
+  async uploadFile(file: FileMetadata): Promise<FileUploadResponse> {
+    try {
+      const fileBuffer = await fs.readFile(file.path)
+      const response = await this.client.files.upload({
+        file: {
+          fileName: file.origin_name,
+          content: new Uint8Array(fileBuffer)
+        },
+        purpose: 'ocr'
+      })
+
+      return {
+        fileId: response.id,
+        displayName: file.origin_name,
+        status: 'success',
+        originalFile: {
+          type: 'mistral',
+          file: response
+        }
+      }
+    } catch (error) {
+      Logger.error('Error uploading file:', error)
+      return {
+        fileId: '',
+        displayName: file.origin_name,
+        status: 'failed'
+      }
+    }
+  }
+
+  async listFiles(): Promise<FileListResponse> {
+    try {
+      const response = await this.client.files.list({})
+      return {
+        files: response.data.map((file) => ({
+          id: file.id,
+          displayName: file.filename || '',
+          size: file.sizeBytes,
+          status: 'success', // All listed files are processed,
+          originalFile: {
+            type: 'mistral',
+            file
+          }
+        }))
+      }
+    } catch (error) {
+      Logger.error('Error listing files:', error)
+      return { files: [] }
+    }
+  }
+
+  async deleteFile(fileId: string): Promise<void> {
+    try {
+      await this.client.files.delete({
+        fileId
+      })
+      Logger.info(`File ${fileId} deleted`)
+    } catch (error) {
+      Logger.error('Error deleting file:', error)
+      throw error
+    }
+  }
+
+  async retrieveFile(fileId: string): Promise<FileUploadResponse> {
+    try {
+      const response = await this.client.files.retrieve({
+        fileId
+      })
+
+      return {
+        fileId: response.id,
+        displayName: response.filename || '',
+        status: 'success' // Retrieved files are always processed
+      }
+    } catch (error) {
+      Logger.error('Error retrieving file:', error)
+      return {
+        fileId: fileId,
+        displayName: '',
+        status: 'failed',
+        originalFile: undefined
+      }
+    }
+  }
+}

src/main/services/urlschema/handle-providers.ts

@@ -1,37 +1,47 @@
-import { IpcChannel } from '@shared/IpcChannel'
 import Logger from 'electron-log'
 
 import { windowService } from '../WindowService'
 
-export function handleProvidersProtocolUrl(url: URL) {
-  const params = new URLSearchParams(url.search)
+export async function handleProvidersProtocolUrl(url: URL) {
   switch (url.pathname) {
     case '/api-keys': {
       // jsonConfig example:
       // {
       //   "id": "tokenflux",
       //   "baseUrl": "https://tokenflux.ai/v1",
-      //   "apiKey": "sk-xxxx"
+      //   "apiKey": "sk-xxxx",
+      //   "name": "TokenFlux", // optional
+      //   "type": "openai" // optional
       // }
-      // cherrystudio://providers/api-keys?data={base64Encode(JSON.stringify(jsonConfig))}
+      // cherrystudio://providers/api-keys?v=1&data={base64Encode(JSON.stringify(jsonConfig))}
+
+      // replace + and / to _ and - because + and / are processed by URLSearchParams
+      const processedSearch = url.search.replaceAll('+', '_').replaceAll('/', '-')
+      const params = new URLSearchParams(processedSearch)
       const data = params.get('data')
-      if (data) {
-        const stringify = Buffer.from(data, 'base64').toString('utf8')
-        Logger.info('get api keys from urlschema: ', stringify)
-        const jsonConfig = JSON.parse(stringify)
-        Logger.info('get api keys from urlschema: ', jsonConfig)
-        const mainWindow = windowService.getMainWindow()
-        if (mainWindow && !mainWindow.isDestroyed()) {
-          mainWindow.webContents.send(IpcChannel.Provider_AddKey, jsonConfig)
-          mainWindow.webContents.executeJavaScript(`window.navigate('/settings/provider?id=${jsonConfig.id}')`)
-        }
+      const mainWindow = windowService.getMainWindow()
+      const version = params.get('v')
+      if (version == '1') {
+        // TODO: handle different version
+        Logger.info('handleProvidersProtocolUrl', { data, version })
+      }
+
+      // add check there is window.navigate function in mainWindow
+      if (
+        mainWindow &&
+        !mainWindow.isDestroyed() &&
+        (await mainWindow.webContents.executeJavaScript(`typeof window.navigate === 'function'`))
+      ) {
+        mainWindow.webContents.executeJavaScript(`window.navigate('/settings/provider?addProviderData=${data}')`)
       } else {
-        Logger.error('No data found in URL')
+        setTimeout(() => {
+          handleProvidersProtocolUrl(url)
+        }, 1000)
       }
       break
     }
     default:
-      console.error(`Unknown MCP protocol URL: ${url}`)
+      Logger.error(`Unknown MCP protocol URL: ${url}`)
       break
   }
 }

src/main/services/urlschema/mcp-install.ts

@@ -65,7 +65,7 @@ export function handleMcpProtocolUrl(url: URL) {
 
       const mainWindow = windowService.getMainWindow()
       if (mainWindow && !mainWindow.isDestroyed()) {
-        mainWindow.webContents.executeJavaScript("window.navigate('/mcp-servers')")
+        mainWindow.webContents.executeJavaScript("window.navigate('/settings/mcp')")
       }
       break
     }

src/main/utils/__tests__/file.test.ts

@@ -92,6 +92,7 @@ describe('file', () => {
     it('should return DOCUMENT for document extensions', () => {
       expect(getFileType('.pdf')).toBe(FileTypes.DOCUMENT)
       expect(getFileType('.pptx')).toBe(FileTypes.DOCUMENT)
+      expect(getFileType('.doc')).toBe(FileTypes.DOCUMENT)
       expect(getFileType('.docx')).toBe(FileTypes.DOCUMENT)
       expect(getFileType('.xlsx')).toBe(FileTypes.DOCUMENT)
       expect(getFileType('.odt')).toBe(FileTypes.DOCUMENT)

src/main/utils/file.ts

@@ -2,12 +2,26 @@ import * as fs from 'node:fs'
 import os from 'node:os'
 import path from 'node:path'
 
-import { isMac } from '@main/constant'
+import { isLinux, isPortable } from '@main/constant'
 import { audioExts, documentExts, imageExts, textExts, videoExts } from '@shared/config/constant'
-import { FileType, FileTypes } from '@types'
+import { FileMetadata, FileTypes } from '@types'
 import { app } from 'electron'
 import { v4 as uuidv4 } from 'uuid'
 
+export function initAppDataDir() {
+  const appDataPath = getAppDataPathFromConfig()
+  if (appDataPath) {
+    app.setPath('userData', appDataPath)
+    return
+  }
+
+  if (isPortable) {
+    const portableDir = process.env.PORTABLE_EXECUTABLE_DIR
+    app.setPath('userData', path.join(portableDir || app.getPath('exe'), 'data'))
+    return
+  }
+}
+
 // 创建文件类型映射表，提高查找效率
 const fileTypeMap = new Map<string, FileTypes>()
 
@@ -23,12 +37,112 @@ function initFileTypeMap() {
 // 初始化映射表
 initFileTypeMap()
 
+export function hasWritePermission(path: string) {
+  try {
+    fs.accessSync(path, fs.constants.W_OK)
+    return true
+  } catch (error) {
+    return false
+  }
+}
+
+function getAppDataPathFromConfig() {
+  try {
+    const configPath = path.join(getConfigDir(), 'config.json')
+    if (!fs.existsSync(configPath)) {
+      return null
+    }
+
+    const config = JSON.parse(fs.readFileSync(configPath, 'utf-8'))
+
+    if (!config.appDataPath) {
+      return null
+    }
+
+    let executablePath = app.getPath('exe')
+    if (isLinux && process.env.APPIMAGE) {
+      // 如果是 AppImage 打包的应用，直接使用 APPIMAGE 环境变量
+      // 这样可以确保获取到正确的可执行文件路径
+      executablePath = path.join(path.dirname(process.env.APPIMAGE), 'cherry-studio.appimage')
+    }
+
+    let appDataPath = null
+    // 兼容旧版本
+    if (config.appDataPath && typeof config.appDataPath === 'string') {
+      appDataPath = config.appDataPath
+      // 将旧版本数据迁移到新版本
+      appDataPath && updateAppDataConfig(appDataPath)
+    } else {
+      appDataPath = config.appDataPath.find(
+        (item: { executablePath: string }) => item.executablePath === executablePath
+      )?.dataPath
+    }
+
+    if (appDataPath && fs.existsSync(appDataPath) && hasWritePermission(appDataPath)) {
+      return appDataPath
+    }
+
+    return null
+  } catch (error) {
+    return null
+  }
+}
+
+export function updateAppDataConfig(appDataPath: string) {
+  const configDir = getConfigDir()
+  if (!fs.existsSync(configDir)) {
+    fs.mkdirSync(configDir, { recursive: true })
+  }
+
+  // config.json
+  // appDataPath: [{ executablePath: string, dataPath: string }]
+  const configPath = path.join(getConfigDir(), 'config.json')
+  let executablePath = app.getPath('exe')
+  if (isLinux && process.env.APPIMAGE) {
+    executablePath = path.join(path.dirname(process.env.APPIMAGE), 'cherry-studio.appimage')
+  }
+
+  if (!fs.existsSync(configPath)) {
+    fs.writeFileSync(configPath, JSON.stringify({ appDataPath: [{ executablePath, dataPath: appDataPath }] }, null, 2))
+    return
+  }
+
+  const config = JSON.parse(fs.readFileSync(configPath, 'utf-8'))
+  if (!config.appDataPath || (config.appDataPath && typeof config.appDataPath !== 'object')) {
+    config.appDataPath = []
+  }
+
+  const existingPath = config.appDataPath.find(
+    (item: { executablePath: string }) => item.executablePath === executablePath
+  )
+
+  if (existingPath) {
+    existingPath.dataPath = appDataPath
+  } else {
+    config.appDataPath.push({ executablePath, dataPath: appDataPath })
+  }
+
+  fs.writeFileSync(configPath, JSON.stringify(config, null, 2))
+}
+
 export function getFileType(ext: string): FileTypes {
   ext = ext.toLowerCase()
   return fileTypeMap.get(ext) || FileTypes.OTHER
 }
 
-export function getAllFiles(dirPath: string, arrayOfFiles: FileType[] = []): FileType[] {
+export function getFileDir(filePath: string) {
+  return path.dirname(filePath)
+}
+
+export function getFileName(filePath: string) {
+  return path.basename(filePath)
+}
+
+export function getFileExt(filePath: string) {
+  return path.extname(filePath)
+}
+
+export function getAllFiles(dirPath: string, arrayOfFiles: FileMetadata[] = []): FileMetadata[] {
   const files = fs.readdirSync(dirPath)
 
   files.forEach((file) => {
@@ -50,7 +164,7 @@ export function getAllFiles(dirPath: string, arrayOfFiles: FileType[] = []): Fil
       const name = path.basename(file)
       const size = fs.statSync(fullPath).size
 
-      const fileItem: FileType = {
+      const fileItem: FileMetadata = {
         id: uuidv4(),
         name,
         path: fullPath,
@@ -88,12 +202,3 @@ export function getCacheDir() {
 export function getAppConfigDir(name: string) {
   return path.join(getConfigDir(), name)
 }
-
-export function setUserDataDir() {
-  if (!isMac) {
-    const dir = path.join(path.dirname(app.getPath('exe')), 'data')
-    if (fs.existsSync(dir) && fs.statSync(dir).isDirectory()) {
-      app.setPath('userData', dir)
-    }
-  }
-}

src/main/utils/process.ts

@@ -49,7 +49,7 @@ export async function getBinaryPath(name?: string): Promise<string> {
 
   const binaryName = await getBinaryName(name)
   const binariesDir = path.join(os.homedir(), '.cherrystudio', 'bin')
-  const binariesDirExists = await fs.existsSync(binariesDir)
+  const binariesDirExists = fs.existsSync(binariesDir)
   return binariesDirExists ? path.join(binariesDir, binaryName) : binaryName
 }
 

src/main/utils/systemInfo.ts

@@ -0,0 +1,92 @@
+import { app } from 'electron'
+import macosRelease from 'macos-release'
+import os from 'os'
+
+/**
+ * System information interface
+ */
+export interface SystemInfo {
+  platform: NodeJS.Platform
+  arch: string
+  osRelease: string
+  appVersion: string
+  osString: string
+  archString: string
+}
+
+/**
+ * Get basic system constants for quick access
+ * @returns {Object} Basic system constants
+ */
+export function getSystemConstants() {
+  return {
+    platform: process.platform,
+    arch: process.arch,
+    osRelease: os.release(),
+    appVersion: app.getVersion()
+  }
+}
+
+/**
+ * Get system information
+ * @returns {SystemInfo} Complete system information object
+ */
+export function getSystemInfo(): SystemInfo {
+  const platform = process.platform
+  const arch = process.arch
+  const osRelease = os.release()
+  const appVersion = app.getVersion()
+
+  let osString = ''
+
+  switch (platform) {
+    case 'win32': {
+      // Get Windows version
+      const parts = osRelease.split('.')
+      const buildNumber = parseInt(parts[2], 10)
+      osString = buildNumber >= 22000 ? 'Windows 11' : 'Windows 10'
+      break
+    }
+    case 'darwin': {
+      // macOS version handling using macos-release for better accuracy
+      try {
+        const macVersionInfo = macosRelease()
+        const versionString = macVersionInfo.version.replace(/\./g, '_') // 15.0.0 -> 15_0_0
+        osString = arch === 'arm64' ? `Mac OS X ${versionString}` : `Intel Mac OS X ${versionString}` // Mac OS X 15_0_0
+      } catch (error) {
+        // Fallback to original logic if macos-release fails
+        const macVersion = osRelease.split('.').slice(0, 2).join('_')
+        osString = arch === 'arm64' ? `Mac OS X ${macVersion}` : `Intel Mac OS X ${macVersion}`
+      }
+      break
+    }
+    case 'linux': {
+      osString = `Linux ${arch}`
+      break
+    }
+    default: {
+      osString = `${platform} ${arch}`
+    }
+  }
+
+  const archString = arch === 'x64' ? 'x86_64' : arch === 'arm64' ? 'arm64' : arch
+
+  return {
+    platform,
+    arch,
+    osRelease,
+    appVersion,
+    osString,
+    archString
+  }
+}
+
+/**
+ * Generate User-Agent string based on user system data
+ * @returns {string} Dynamically generated User-Agent string
+ */
+export function generateUserAgent(): string {
+  const systemInfo = getSystemInfo()
+
+  return `Mozilla/5.0 (${systemInfo.osString}; ${systemInfo.archString}) AppleWebKit/537.36 (KHTML, like Gecko) CherryStudio/${systemInfo.appVersion} Chrome/124.0.0.0 Safari/537.36`
+}

src/preload/index.ts

@@ -1,8 +1,19 @@
 import type { ExtractChunkData } from '@cherrystudio/embedjs-interfaces'
 import { electronAPI } from '@electron-toolkit/preload'
-import { FeedUrl } from '@shared/config/constant'
+import { UpgradeChannel } from '@shared/config/constant'
 import { IpcChannel } from '@shared/IpcChannel'
-import { FileType, KnowledgeBaseParams, KnowledgeItem, MCPServer, Shortcut, ThemeMode, WebDavConfig } from '@types'
+import {
+  FileListResponse,
+  FileMetadata,
+  FileUploadResponse,
+  KnowledgeBaseParams,
+  KnowledgeItem,
+  MCPServer,
+  Provider,
+  Shortcut,
+  ThemeMode,
+  WebDavConfig
+} from '@types'
 import { contextBridge, ipcRenderer, OpenDialogOptions, shell, webUtils } from 'electron'
 import { Notification } from 'src/renderer/src/types/notification'
 import { CreateDirectoryOptions } from 'webdav'
@@ -17,18 +28,35 @@ const api = {
   checkForUpdate: () => ipcRenderer.invoke(IpcChannel.App_CheckForUpdate),
   showUpdateDialog: () => ipcRenderer.invoke(IpcChannel.App_ShowUpdateDialog),
   setLanguage: (lang: string) => ipcRenderer.invoke(IpcChannel.App_SetLanguage, lang),
+  setEnableSpellCheck: (isEnable: boolean) => ipcRenderer.invoke(IpcChannel.App_SetEnableSpellCheck, isEnable),
+  setSpellCheckLanguages: (languages: string[]) => ipcRenderer.invoke(IpcChannel.App_SetSpellCheckLanguages, languages),
   setLaunchOnBoot: (isActive: boolean) => ipcRenderer.invoke(IpcChannel.App_SetLaunchOnBoot, isActive),
   setLaunchToTray: (isActive: boolean) => ipcRenderer.invoke(IpcChannel.App_SetLaunchToTray, isActive),
   setTray: (isActive: boolean) => ipcRenderer.invoke(IpcChannel.App_SetTray, isActive),
   setTrayOnClose: (isActive: boolean) => ipcRenderer.invoke(IpcChannel.App_SetTrayOnClose, isActive),
-  setFeedUrl: (feedUrl: FeedUrl) => ipcRenderer.invoke(IpcChannel.App_SetFeedUrl, feedUrl),
+  setTestPlan: (isActive: boolean) => ipcRenderer.invoke(IpcChannel.App_SetTestPlan, isActive),
+  setTestChannel: (channel: UpgradeChannel) => ipcRenderer.invoke(IpcChannel.App_SetTestChannel, channel),
   setTheme: (theme: ThemeMode) => ipcRenderer.invoke(IpcChannel.App_SetTheme, theme),
   handleZoomFactor: (delta: number, reset: boolean = false) =>
     ipcRenderer.invoke(IpcChannel.App_HandleZoomFactor, delta, reset),
   setAutoUpdate: (isActive: boolean) => ipcRenderer.invoke(IpcChannel.App_SetAutoUpdate, isActive),
+  select: (options: Electron.OpenDialogOptions) => ipcRenderer.invoke(IpcChannel.App_Select, options),
+  hasWritePermission: (path: string) => ipcRenderer.invoke(IpcChannel.App_HasWritePermission, path),
+  setAppDataPath: (path: string) => ipcRenderer.invoke(IpcChannel.App_SetAppDataPath, path),
+  getDataPathFromArgs: () => ipcRenderer.invoke(IpcChannel.App_GetDataPathFromArgs),
+  copy: (oldPath: string, newPath: string, occupiedDirs: string[] = []) =>
+    ipcRenderer.invoke(IpcChannel.App_Copy, oldPath, newPath, occupiedDirs),
+  setStopQuitApp: (stop: boolean, reason: string) => ipcRenderer.invoke(IpcChannel.App_SetStopQuitApp, stop, reason),
+  flushAppData: () => ipcRenderer.invoke(IpcChannel.App_FlushAppData),
+  isNotEmptyDir: (path: string) => ipcRenderer.invoke(IpcChannel.App_IsNotEmptyDir, path),
+  relaunchApp: (options?: Electron.RelaunchOptions) => ipcRenderer.invoke(IpcChannel.App_RelaunchApp, options),
   openWebsite: (url: string) => ipcRenderer.invoke(IpcChannel.Open_Website, url),
   getCacheSize: () => ipcRenderer.invoke(IpcChannel.App_GetCacheSize),
   clearCache: () => ipcRenderer.invoke(IpcChannel.App_ClearCache),
+  mac: {
+    isProcessTrusted: (): Promise<boolean> => ipcRenderer.invoke(IpcChannel.App_MacIsProcessTrusted),
+    requestProcessTrust: (): Promise<boolean> => ipcRenderer.invoke(IpcChannel.App_MacRequestProcessTrust)
+  },
   notification: {
     send: (notification: Notification) => ipcRenderer.invoke(IpcChannel.Notification_Send, notification)
   },
@@ -62,13 +90,25 @@ const api = {
   },
   file: {
     select: (options?: OpenDialogOptions) => ipcRenderer.invoke(IpcChannel.File_Select, options),
-    upload: (file: FileType) => ipcRenderer.invoke(IpcChannel.File_Upload, file),
+    upload: (file: FileMetadata) => ipcRenderer.invoke(IpcChannel.File_Upload, file),
     delete: (fileId: string) => ipcRenderer.invoke(IpcChannel.File_Delete, fileId),
+    deleteDir: (dirPath: string) => ipcRenderer.invoke(IpcChannel.File_DeleteDir, dirPath),
     read: (fileId: string) => ipcRenderer.invoke(IpcChannel.File_Read, fileId),
     clear: () => ipcRenderer.invoke(IpcChannel.File_Clear),
     get: (filePath: string) => ipcRenderer.invoke(IpcChannel.File_Get, filePath),
-    create: (fileName: string) => ipcRenderer.invoke(IpcChannel.File_Create, fileName),
+    /**
+     * 创建一个空的临时文件
+     * @param fileName 文件名
+     * @returns 临时文件路径
+     */
+    createTempFile: (fileName: string): Promise<string> => ipcRenderer.invoke(IpcChannel.File_CreateTempFile, fileName),
+    /**
+     * 写入文件
+     * @param filePath 文件路径
+     * @param data 数据
+     */
     write: (filePath: string, data: Uint8Array | string) => ipcRenderer.invoke(IpcChannel.File_Write, filePath, data),
+
     writeWithId: (id: string, content: string) => ipcRenderer.invoke(IpcChannel.File_WriteWithId, id, content),
     open: (options?: OpenDialogOptions) => ipcRenderer.invoke(IpcChannel.File_Open, options),
     openPath: (path: string) => ipcRenderer.invoke(IpcChannel.File_OpenPath, path),
@@ -76,13 +116,14 @@ const api = {
       ipcRenderer.invoke(IpcChannel.File_Save, path, content, options),
     selectFolder: () => ipcRenderer.invoke(IpcChannel.File_SelectFolder),
     saveImage: (name: string, data: string) => ipcRenderer.invoke(IpcChannel.File_SaveImage, name, data),
+    binaryImage: (fileId: string) => ipcRenderer.invoke(IpcChannel.File_BinaryImage, fileId),
     base64Image: (fileId: string) => ipcRenderer.invoke(IpcChannel.File_Base64Image, fileId),
     saveBase64Image: (data: string) => ipcRenderer.invoke(IpcChannel.File_SaveBase64Image, data),
     download: (url: string, isUseContentType?: boolean) =>
       ipcRenderer.invoke(IpcChannel.File_Download, url, isUseContentType),
     copy: (fileId: string, destPath: string) => ipcRenderer.invoke(IpcChannel.File_Copy, fileId, destPath),
-    binaryImage: (fileId: string) => ipcRenderer.invoke(IpcChannel.File_BinaryImage, fileId),
     base64File: (fileId: string) => ipcRenderer.invoke(IpcChannel.File_Base64File, fileId),
+    pdfInfo: (fileId: string) => ipcRenderer.invoke(IpcChannel.File_GetPdfInfo, fileId),
     getPathForFile: (file: File) => webUtils.getPathForFile(file)
   },
   fs: {
@@ -102,31 +143,45 @@ const api = {
     add: ({
       base,
       item,
+      userId,
       forceReload = false
     }: {
       base: KnowledgeBaseParams
       item: KnowledgeItem
+      userId?: string
       forceReload?: boolean
-    }) => ipcRenderer.invoke(IpcChannel.KnowledgeBase_Add, { base, item, forceReload }),
+    }) => ipcRenderer.invoke(IpcChannel.KnowledgeBase_Add, { base, item, forceReload, userId }),
     remove: ({ uniqueId, uniqueIds, base }: { uniqueId: string; uniqueIds: string[]; base: KnowledgeBaseParams }) =>
       ipcRenderer.invoke(IpcChannel.KnowledgeBase_Remove, { uniqueId, uniqueIds, base }),
     search: ({ search, base }: { search: string; base: KnowledgeBaseParams }) =>
       ipcRenderer.invoke(IpcChannel.KnowledgeBase_Search, { search, base }),
     rerank: ({ search, base, results }: { search: string; base: KnowledgeBaseParams; results: ExtractChunkData[] }) =>
-      ipcRenderer.invoke(IpcChannel.KnowledgeBase_Rerank, { search, base, results })
+      ipcRenderer.invoke(IpcChannel.KnowledgeBase_Rerank, { search, base, results }),
+    checkQuota: ({ base, userId }: { base: KnowledgeBaseParams; userId: string }) =>
+      ipcRenderer.invoke(IpcChannel.KnowledgeBase_Check_Quota, base, userId)
   },
   window: {
     setMinimumSize: (width: number, height: number) =>
       ipcRenderer.invoke(IpcChannel.Windows_SetMinimumSize, width, height),
     resetMinimumSize: () => ipcRenderer.invoke(IpcChannel.Windows_ResetMinimumSize)
   },
-  gemini: {
-    uploadFile: (file: FileType, { apiKey, baseURL }: { apiKey: string; baseURL: string }) =>
-      ipcRenderer.invoke(IpcChannel.Gemini_UploadFile, file, { apiKey, baseURL }),
-    base64File: (file: FileType) => ipcRenderer.invoke(IpcChannel.Gemini_Base64File, file),
-    retrieveFile: (file: FileType, apiKey: string) => ipcRenderer.invoke(IpcChannel.Gemini_RetrieveFile, file, apiKey),
-    listFiles: (apiKey: string) => ipcRenderer.invoke(IpcChannel.Gemini_ListFiles, apiKey),
-    deleteFile: (fileId: string, apiKey: string) => ipcRenderer.invoke(IpcChannel.Gemini_DeleteFile, fileId, apiKey)
+  fileService: {
+    upload: (provider: Provider, file: FileMetadata): Promise<FileUploadResponse> =>
+      ipcRenderer.invoke(IpcChannel.FileService_Upload, provider, file),
+    list: (provider: Provider): Promise<FileListResponse> => ipcRenderer.invoke(IpcChannel.FileService_List, provider),
+    delete: (provider: Provider, fileId: string) => ipcRenderer.invoke(IpcChannel.FileService_Delete, provider, fileId),
+    retrieve: (provider: Provider, fileId: string): Promise<FileUploadResponse> =>
+      ipcRenderer.invoke(IpcChannel.FileService_Retrieve, provider, fileId)
+  },
+  selectionMenu: {
+    action: (action: string) => ipcRenderer.invoke('selection-menu:action', action)
+  },
+
+  vertexAI: {
+    getAuthHeaders: (params: { projectId: string; serviceAccount?: { privateKey: string; clientEmail: string } }) =>
+      ipcRenderer.invoke(IpcChannel.VertexAI_GetAuthHeaders, params),
+    clearAuthCache: (projectId: string, clientEmail?: string) =>
+      ipcRenderer.invoke(IpcChannel.VertexAI_ClearAuthCache, projectId, clientEmail)
   },
   config: {
     set: (key: string, value: any, isNotify: boolean = false) =>
@@ -162,6 +217,10 @@ const api = {
     getInstallInfo: () => ipcRenderer.invoke(IpcChannel.Mcp_GetInstallInfo),
     checkMcpConnectivity: (server: any) => ipcRenderer.invoke(IpcChannel.Mcp_CheckConnectivity, server)
   },
+  python: {
+    execute: (script: string, context?: Record<string, any>, timeout?: number) =>
+      ipcRenderer.invoke(IpcChannel.Python_Execute, script, context, timeout)
+  },
   shell: {
     openExternal: (url: string, options?: Electron.OpenExternalOptions) => shell.openExternal(url, options)
   },
@@ -204,7 +263,9 @@ const api = {
   },
   webview: {
     setOpenLinkExternal: (webviewId: number, isExternal: boolean) =>
-      ipcRenderer.invoke(IpcChannel.Webview_SetOpenLinkExternal, webviewId, isExternal)
+      ipcRenderer.invoke(IpcChannel.Webview_SetOpenLinkExternal, webviewId, isExternal),
+    setSpellCheckEnabled: (webviewId: number, isEnable: boolean) =>
+      ipcRenderer.invoke(IpcChannel.Webview_SetSpellCheckEnabled, webviewId, isEnable)
   },
   storeSync: {
     subscribe: () => ipcRenderer.invoke(IpcChannel.StoreSync_Subscribe),

src/renderer/selectionToolbar.html

@@ -2,42 +2,45 @@
 <html lang="zh-CN">
 
 <head>
-    <meta charset="UTF-8" />
-    <meta name="viewport" content="initial-scale=1, width=device-width" />
-    <meta http-equiv="Content-Security-Policy"
-        content="default-src 'self'; connect-src blob: *; script-src 'self' 'unsafe-eval' *; worker-src 'self' blob:; style-src 'self' 'unsafe-inline' *; font-src 'self' data: *; img-src 'self' data: file: * blob:; frame-src * file:" />
-    <title>Cherry Studio Selection Toolbar</title>
+  <meta charset="UTF-8" />
+  <meta name="viewport" content="initial-scale=1, width=device-width" />
+  <meta http-equiv="Content-Security-Policy"
+    content="default-src 'self'; connect-src blob: *; script-src 'self' 'unsafe-eval' *; worker-src 'self' blob:; style-src 'self' 'unsafe-inline' *; font-src 'self' data: *; img-src 'self' data: file: * blob:; frame-src * file:" />
+  <title>Cherry Studio Selection Toolbar</title>
 
 </head>
 
 <body>
-    <div id="root"></div>
-    <script type="module" src="/src/windows/selection/toolbar/entryPoint.tsx"></script>
-    <style>
-        html {
-            margin: 0;
-        }
+  <div id="root"></div>
+  <script type="module" src="/src/windows/selection/toolbar/entryPoint.tsx"></script>
+  <style>
+    html {
+      margin: 0 !important;
+      background-color: transparent !important;
+      background-image: none !important;
 
-        body {
-            margin: 0;
-            padding: 0;
-            overflow: hidden;
-            width: 100vw;
-            height: 100vh;
+    }
 
-            -webkit-user-select: none;
-            -moz-user-select: none;
-            -ms-user-select: none;
-            user-select: none;
-        }
+    body {
+      margin: 0 !important;
+      padding: 0 !important;
+      overflow: hidden !important;
+      width: 100vw !important;
+      height: 100vh !important;
 
-        #root {
-            margin: 0;
-            padding: 0;
-            width: max-content !important;
-            height: fit-content !important;
-        }
-    </style>
+      -webkit-user-select: none;
+      -moz-user-select: none;
+      -ms-user-select: none;
+      user-select: none;
+    }
+
+    #root {
+      margin: 0 !important;
+      padding: 0 !important;
+      width: max-content !important;
+      height: fit-content !important;
+    }
+  </style>
 </body>
 
 </html>

src/renderer/src/App.tsx

@@ -5,7 +5,7 @@ import { Provider } from 'react-redux'
 import { HashRouter, Route, Routes } from 'react-router-dom'
 import { PersistGate } from 'redux-persist/integration/react'
 
-import MainSidebar from './components/app/MainSidebar'
+import Sidebar from './components/app/Sidebar'
 import TopViewContainer from './components/TopView'
 import AntdProvider from './context/AntdProvider'
 import { CodeStyleProvider } from './context/CodeStyleProvider'
@@ -13,9 +13,14 @@ import { NotificationProvider } from './context/NotificationProvider'
 import StyleSheetManager from './context/StyleSheetManager'
 import { ThemeProvider } from './context/ThemeProvider'
 import NavigationHandler from './handler/NavigationHandler'
-import DiscoverPage from './pages/discover'
+import AgentsPage from './pages/agents/AgentsPage'
+import AppsPage from './pages/apps/AppsPage'
+import FilesPage from './pages/files/FilesPage'
 import HomePage from './pages/home/HomePage'
+import KnowledgePage from './pages/knowledge/KnowledgePage'
+import PaintingsRoutePage from './pages/paintings/PaintingsRoutePage'
 import SettingsPage from './pages/settings/SettingsPage'
+import TranslatePage from './pages/translate/TranslatePage'
 
 function App(): React.ReactElement {
   return (
@@ -29,18 +34,16 @@ function App(): React.ReactElement {
                   <TopViewContainer>
                     <HashRouter>
                       <NavigationHandler />
-                      <MainSidebar />
+                      <Sidebar />
                       <Routes>
                         <Route path="/" element={<HomePage />} />
-                        {/* <Route path="/agents" element={<AgentsPage />} /> */}
-                        {/* <Route path="/paintings/*" element={<PaintingsRoutePage />} /> */}
-                        {/* <Route path="/translate" element={<TranslatePage />} /> */}
-                        {/* <Route path="/files" element={<FilesPage />} /> */}
-                        {/* <Route path="/knowledge" element={<KnowledgePage />} /> */}
-                        {/* <Route path="/apps" element={<AppsPage />} /> */}
-                        {/* <Route path="/mcp-servers/*" element={<McpServersPage />} /> */}
+                        <Route path="/agents" element={<AgentsPage />} />
+                        <Route path="/paintings/*" element={<PaintingsRoutePage />} />
+                        <Route path="/translate" element={<TranslatePage />} />
+                        <Route path="/files" element={<FilesPage />} />
+                        <Route path="/knowledge" element={<KnowledgePage />} />
+                        <Route path="/apps" element={<AppsPage />} />
                         <Route path="/settings/*" element={<SettingsPage />} />
-                        <Route path="/discover/*" element={<DiscoverPage />} />
                       </Routes>
                     </HashRouter>
                   </TopViewContainer>

src/renderer/src/aiCore/AI_CORE_DESIGN.md

@@ -0,0 +1,223 @@
+# Cherry Studio AI Provider 技术架构文档 (新方案)
+
+## 1. 核心设计理念与目标
+
+本架构旨在重构 Cherry Studio 的 AI Provider（现称为 `aiCore`）层，以实现以下目标：
+
+- **职责清晰**：明确划分各组件的职责，降低耦合度。
+- **高度复用**：最大化业务逻辑和通用处理逻辑的复用，减少重复代码。
+- **易于扩展**：方便快捷地接入新的 AI Provider (LLM供应商) 和添加新的 AI 功能 (如翻译、摘要、图像生成等)。
+- **易于维护**：简化单个组件的复杂性，提高代码的可读性和可维护性。
+- **标准化**：统一内部数据流和接口，简化不同 Provider 之间的差异处理。
+
+核心思路是将纯粹的 **SDK 适配层 (`XxxApiClient`)**、**通用逻辑处理与智能解析层 (中间件)** 以及 **统一业务功能入口层 (`AiCoreService`)** 清晰地分离开来。
+
+## 2. 核心组件详解
+
+### 2.1. `aiCore` (原 `AiProvider` 文件夹)
+
+这是整个 AI 功能的核心模块。
+
+#### 2.1.1. `XxxApiClient` (例如 `aiCore/clients/openai/OpenAIApiClient.ts`)
+
+- **职责**：作为特定 AI Provider SDK 的纯粹适配层。
+  - **参数适配**：将应用内部统一的 `CoreRequest` 对象 (见下文) 转换为特定 SDK 所需的请求参数格式。
+  - **基础响应转换**：将 SDK 返回的原始数据块 (`RawSdkChunk`，例如 `OpenAI.Chat.Completions.ChatCompletionChunk`) 转换为一组最基础、最直接的应用层 `Chunk` 对象 (定义于 `src/renderer/src/types/chunk.ts`)。
+    - 例如：SDK 的 `delta.content` -> `TextDeltaChunk`；SDK 的 `delta.reasoning_content` -> `ThinkingDeltaChunk`；SDK 的 `delta.tool_calls` -> `RawToolCallChunk` (包含原始工具调用数据)。
+    - **关键**：`XxxApiClient` **不处理**耦合在文本内容中的复杂结构，如 `<think>` 或 `<tool_use>` 标签。
+- **特点**：极度轻量化，代码量少，易于实现和维护新的 Provider 适配。
+
+#### 2.1.2. `ApiClient.ts` (或 `BaseApiClient.ts` 的核心接口)
+
+- 定义了所有 `XxxApiClient` 必须实现的接口，如：
+  - `getSdkInstance(): Promise<TSdkInstance> | TSdkInstance`
+  - `getRequestTransformer(): RequestTransformer<TSdkParams>`
+  - `getResponseChunkTransformer(): ResponseChunkTransformer<TRawChunk, TResponseContext>`
+  - 其他可选的、与特定 Provider 相关的辅助方法 (如工具调用转换)。
+
+#### 2.1.3. `ApiClientFactory.ts`
+
+- 根据 Provider 配置动态创建和返回相应的 `XxxApiClient` 实例。
+
+#### 2.1.4. `AiCoreService.ts` (`aiCore/index.ts`)
+
+- **职责**：作为所有 AI 相关业务功能的统一入口。
+  - 提供面向应用的高层接口，例如：
+    - `executeCompletions(params: CompletionsParams): Promise<AggregatedCompletionsResult>`
+    - `translateText(params: TranslateParams): Promise<AggregatedTranslateResult>`
+    - `summarizeText(params: SummarizeParams): Promise<AggregatedSummarizeResult>`
+    - 未来可能的 `generateImage(prompt: string): Promise<ImageResult>` 等。
+  - **返回 `Promise`**：每个服务方法返回一个 `Promise`，该 `Promise` 会在整个（可能是流式的）操作完成后，以包含所有聚合结果（如完整文本、工具调用详情、最终的`usage`/`metrics`等）的对象来 `resolve`。
+  - **支持流式回调**：服务方法的参数 (如 `CompletionsParams`) 依然包含 `onChunk` 回调，用于向调用方实时推送处理过程中的 `Chunk` 数据，实现流式UI更新。
+  - **封装特定任务的提示工程 (Prompt Engineering)**：
+    - 例如，`translateText` 方法内部会构建一个包含特定翻译指令的 `CoreRequest`。
+  - **编排和调用中间件链**：通过内部的 `MiddlewareBuilder` (参见 `middleware/BUILDER_USAGE.md`) 实例，根据调用的业务方法和参数，动态构建和组织合适的中间件序列，然后通过 `applyCompletionsMiddlewares` 等组合函数执行。
+  - 获取 `ApiClient` 实例并将其注入到中间件上游的 `Context` 中。
+  - **将 `Promise` 的 `resolve` 和 `reject` 函数传递给中间件链** (通过 `Context`)，以便 `FinalChunkConsumerAndNotifierMiddleware` 可以在操作完成或发生错误时结束该 `Promise`。
+- **优势**：
+  - 业务逻辑（如翻译、摘要的提示构建和流程控制）只需实现一次，即可支持所有通过 `ApiClient` 接入的底层 Provider。
+  - **支持外部编排**：调用方可以 `await` 服务方法以获取最终聚合结果，然后将此结果作为后续操作的输入，轻松实现多步骤工作流。
+  - **支持内部组合**：服务自身也可以通过 `await` 调用其他原子服务方法来构建更复杂的组合功能。
+
+#### 2.1.5. `coreRequestTypes.ts` (或 `types.ts`)
+
+- 定义核心的、Provider 无关的内部请求结构，例如：
+  - `CoreCompletionsRequest`: 包含标准化后的消息列表、模型配置、工具列表、最大Token数、是否流式输出等。
+  - `CoreTranslateRequest`, `CoreSummarizeRequest` 等 (如果与 `CoreCompletionsRequest` 结构差异较大，否则可复用并添加任务类型标记)。
+
+### 2.2. `middleware`
+
+中间件层负责处理请求和响应流中的通用逻辑和特定特性。其设计和使用遵循 `middleware/BUILDER_USAGE.md` 中定义的规范。
+
+**核心组件包括：**
+
+- **`MiddlewareBuilder`**: 一个通用的、提供流式API的类，用于动态构建中间件链。它支持从基础链开始，根据条件添加、插入、替换或移除中间件。
+- **`applyCompletionsMiddlewares`**: 负责接收 `MiddlewareBuilder` 构建的链并按顺序执行，专门用于 Completions 流程。
+- **`MiddlewareRegistry`**: 集中管理所有可用中间件的注册表，提供统一的中间件访问接口。
+- **各种独立的中间件模块** (存放于 `common/`, `core/`, `feat/` 子目录)。
+
+#### 2.2.1. `middlewareTypes.ts`
+
+- 定义中间件的核心类型，如 `AiProviderMiddlewareContext` (扩展后包含 `_apiClientInstance` 和 `_coreRequest`)、`MiddlewareAPI`、`CompletionsMiddleware` 等。
+
+#### 2.2.2. 核心中间件 (`middleware/core/`)
+
+- **`TransformCoreToSdkParamsMiddleware.ts`**: 调用 `ApiClient.getRequestTransformer()` 将 `CoreRequest` 转换为特定 SDK 的参数，并存入上下文。
+- **`RequestExecutionMiddleware.ts`**: 调用 `ApiClient.getSdkInstance()` 获取 SDK 实例，并使用转换后的参数执行实际的 API 调用，返回原始 SDK 流。
+- **`StreamAdapterMiddleware.ts`**: 将各种形态的原始 SDK 流 (如异步迭代器) 统一适配为 `ReadableStream<RawSdkChunk>`。
+  - **`RawSdkChunk`**：指特定AI提供商SDK在流式响应中返回的、未经应用层统一处理的原始数据块格式 (例如 OpenAI 的 `ChatCompletionChunk`，Gemini 的 `GenerateContentResponse` 中的部分等)。
+- **`RawSdkChunkToAppChunkMiddleware.ts`**: (新增) 消费 `ReadableStream<RawSdkChunk>`，在其内部对每个 `RawSdkChunk` 调用 `ApiClient.getResponseChunkTransformer()`，将其转换为一个或多个基础的应用层 `Chunk` 对象，并输出 `ReadableStream<Chunk>`。
+
+#### 2.2.3. 特性中间件 (`middleware/feat/`)
+
+这些中间件消费由 `ResponseTransformMiddleware` 输出的、相对标准化的 `Chunk` 流，并处理更复杂的逻辑。
+
+- **`ThinkingTagExtractionMiddleware.ts`**: 检查 `TextDeltaChunk`，解析其中可能包含的 `<think>...</think>` 文本内嵌标签，生成 `ThinkingDeltaChunk` 和 `ThinkingCompleteChunk`。
+- **`ToolUseExtractionMiddleware.ts`**: 检查 `TextDeltaChunk`，解析其中可能包含的 `<tool_use>...</tool_use>` 文本内嵌标签，生成工具调用相关的 Chunk。如果 `ApiClient` 输出了原生工具调用数据，此中间件也负责将其转换为标准格式。
+
+#### 2.2.4. 核心处理中间件 (`middleware/core/`)
+
+- **`TransformCoreToSdkParamsMiddleware.ts`**: 调用 `ApiClient.getRequestTransformer()` 将 `CoreRequest` 转换为特定 SDK 的参数，并存入上下文。
+- **`SdkCallMiddleware.ts`**: 调用 `ApiClient.getSdkInstance()` 获取 SDK 实例，并使用转换后的参数执行实际的 API 调用，返回原始 SDK 流。
+- **`StreamAdapterMiddleware.ts`**: 将各种形态的原始 SDK 流统一适配为标准流格式。
+- **`ResponseTransformMiddleware.ts`**: 将原始 SDK 响应转换为应用层标准 `Chunk` 对象。
+- **`TextChunkMiddleware.ts`**: 处理文本相关的 Chunk 流。
+- **`ThinkChunkMiddleware.ts`**: 处理思考相关的 Chunk 流。
+- **`McpToolChunkMiddleware.ts`**: 处理工具调用相关的 Chunk 流。
+- **`WebSearchMiddleware.ts`**: 处理 Web 搜索相关逻辑。
+
+#### 2.2.5. 通用中间件 (`middleware/common/`)
+
+- **`LoggingMiddleware.ts`**: 请求和响应日志。
+- **`AbortHandlerMiddleware.ts`**: 处理请求中止。
+- **`FinalChunkConsumerMiddleware.ts`**: 消费最终的 `Chunk` 流，通过 `context.onChunk` 回调通知应用层实时数据。
+  - **累积数据**：在流式处理过程中，累积关键数据，如文本片段、工具调用信息、`usage`/`metrics` 等。
+  - **结束 `Promise`**：当输入流结束时，使用累积的聚合结果来完成整个处理流程。
+  - 在流结束时，发送包含最终累加信息的完成信号。
+
+### 2.3. `types/chunk.ts`
+
+- 定义应用全局统一的 `Chunk` 类型及其所有变体。这包括基础类型 (如 `TextDeltaChunk`, `ThinkingDeltaChunk`)、SDK原生数据传递类型 (如 `RawToolCallChunk`, `RawFinishChunk` - 作为 `ApiClient` 转换的中间产物)，以及功能性类型 (如 `McpToolCallRequestChunk`, `WebSearchCompleteChunk`)。
+
+## 3. 核心执行流程 (以 `AiCoreService.executeCompletions` 为例)
+
+```markdown
+**应用层 (例如 UI 组件)**
+||
+\\/
+**`AiProvider.completions` (`aiCore/index.ts`)**
+(1. prepare ApiClient instance. 2. use `CompletionsMiddlewareBuilder.withDefaults()` to build middleware chain. 3. call `applyCompletionsMiddlewares`)
+||
+\\/
+**`applyCompletionsMiddlewares` (`middleware/composer.ts`)**
+(接收构建好的链、ApiClient实例、原始SDK方法，开始按序执行中间件)
+||
+\\/
+**[ 预处理阶段中间件 ]**
+(例如: `FinalChunkConsumerMiddleware`, `TransformCoreToSdkParamsMiddleware`, `AbortHandlerMiddleware`)
+|| (Context 中准备好 SDK 请求参数)
+\\/
+**[ 处理阶段中间件 ]**
+(例如: `McpToolChunkMiddleware`, `WebSearchMiddleware`, `TextChunkMiddleware`, `ThinkingTagExtractionMiddleware`)
+|| (处理各种特性和Chunk类型)
+\\/
+**[ SDK调用阶段中间件 ]**
+(例如: `ResponseTransformMiddleware`, `StreamAdapterMiddleware`, `SdkCallMiddleware`)
+|| (输出: 标准化的应用层Chunk流)
+\\/
+**`FinalChunkConsumerMiddleware` (核心)**
+(消费最终的 `Chunk` 流, 通过 `context.onChunk` 回调通知应用层, 并在流结束时完成处理)
+||
+\\/
+**`AiProvider.completions` 返回 `Promise<CompletionsResult>`**
+```
+
+## 4. 建议的文件/目录结构
+
+```
+src/renderer/src/
+└── aiCore/
+    ├── clients/
+    │   ├── openai/
+    │   ├── gemini/
+    │   ├── anthropic/
+    │   ├── BaseApiClient.ts
+    │   ├── ApiClientFactory.ts
+    │   ├── AihubmixAPIClient.ts
+    │   ├── index.ts
+    │   └── types.ts
+    ├── middleware/
+    │   ├── common/
+    │   ├── core/
+    │   ├── feat/
+    │   ├── builder.ts
+    │   ├── composer.ts
+    │   ├── index.ts
+    │   ├── register.ts
+    │   ├── schemas.ts
+    │   ├── types.ts
+    │   └── utils.ts
+    ├── types/
+    │   ├── chunk.ts
+    │   └── ...
+    └── index.ts
+```
+
+## 5. 迁移和实施建议
+
+- **小步快跑，逐步迭代**：优先完成核心流程的重构（例如 `completions`），再逐步迁移其他功能（`translate` 等）和其他 Provider。
+- **优先定义核心类型**：`CoreRequest`, `Chunk`, `ApiClient` 接口是整个架构的基石。
+- **为 `ApiClient` 瘦身**：将现有 `XxxProvider` 中的复杂逻辑剥离到新的中间件或 `AiCoreService` 中。
+- **强化中间件**：让中间件承担起更多解析和特性处理的责任。
+- **编写单元测试和集成测试**：确保每个组件和整体流程的正确性。
+
+此架构旨在提供一个更健壮、更灵活、更易于维护的 AI 功能核心，支撑 Cherry Studio 未来的发展。
+
+## 6. 迁移策略与实施建议
+
+本节内容提炼自早期的 `migrate.md` 文档，并根据最新的架构讨论进行了调整。
+
+**目标架构核心组件回顾：**
+
+与第 2 节描述的核心组件一致，主要包括 `XxxApiClient`, `AiCoreService`, 中间件链, `CoreRequest` 类型, 和标准化的 `Chunk` 类型。
+
+**迁移步骤：**
+
+**Phase 0: 准备工作和类型定义**
+
+1.  **定义核心数据结构 (TypeScript 类型)：**
+    - `CoreCompletionsRequest` (Type)：定义应用内部统一的对话请求结构。
+    - `Chunk` (Type - 检查并按需扩展现有 `src/renderer/src/types/chunk.ts`)：定义所有可能的通用Chunk类型。
+    - 为其他API（翻译、总结）定义类似的 `CoreXxxRequest` (Type)。
+2.  **定义 `ApiClient` 接口：** 明确 `getRequestTransformer`, `getResponseChunkTransformer`, `getSdkInstance` 等核心方法。
+3.  **调整 `AiProviderMiddlewareContext`：**
+    - 确保包含 `_apiClientInstance: ApiClient<any,any,any>`。
+    - 确保包含 `_coreRequest: CoreRequestType`。
+    - 考虑添加 `resolvePromise: (value: AggregatedResultType) => void` 和 `rejectPromise: (reason?: any) => void` 用于 `AiCoreService` 的 Promise 返回。
+
+**Phase 1: 实现第一个 `ApiClient` (以 `OpenAIApiClient` 为例)**
+
+1.  **创建 `OpenAIApiClient` 类：** 实现 `ApiClient` 接口。
+2.  **迁移SDK实例和配置。**
+3.  **实现 `getRequestTransformer()`：** 将 `CoreCompletionsRequest` 转换为 OpenAI SDK 参数。
+4.  **实现 `getResponseChunkTransformer()`：** 将 `OpenAI.Chat.Completions.ChatCompletionChunk` 转换为基础的 `

src/renderer/src/aiCore/clients/AihubmixAPIClient.ts

@@ -0,0 +1,223 @@
+import { isOpenAILLMModel } from '@renderer/config/models'
+import {
+  GenerateImageParams,
+  MCPCallToolResponse,
+  MCPTool,
+  MCPToolResponse,
+  Model,
+  Provider,
+  ToolCallResponse
+} from '@renderer/types'
+import {
+  RequestOptions,
+  SdkInstance,
+  SdkMessageParam,
+  SdkModel,
+  SdkParams,
+  SdkRawChunk,
+  SdkRawOutput,
+  SdkTool,
+  SdkToolCall
+} from '@renderer/types/sdk'
+
+import { CompletionsContext } from '../middleware/types'
+import { AnthropicAPIClient } from './anthropic/AnthropicAPIClient'
+import { BaseApiClient } from './BaseApiClient'
+import { GeminiAPIClient } from './gemini/GeminiAPIClient'
+import { OpenAIAPIClient } from './openai/OpenAIApiClient'
+import { OpenAIResponseAPIClient } from './openai/OpenAIResponseAPIClient'
+import { RequestTransformer, ResponseChunkTransformer } from './types'
+
+/**
+ * AihubmixAPIClient - 根据模型类型自动选择合适的ApiClient
+ * 使用装饰器模式实现，在ApiClient层面进行模型路由
+ */
+export class AihubmixAPIClient extends BaseApiClient {
+  // 使用联合类型而不是any，保持类型安全
+  private clients: Map<string, AnthropicAPIClient | GeminiAPIClient | OpenAIResponseAPIClient | OpenAIAPIClient> =
+    new Map()
+  private defaultClient: OpenAIAPIClient
+  private currentClient: BaseApiClient
+
+  constructor(provider: Provider) {
+    super(provider)
+
+    const providerExtraHeaders = {
+      ...provider,
+      extra_headers: {
+        ...provider.extra_headers,
+        'APP-Code': 'MLTG2087'
+      }
+    }
+
+    // 初始化各个client - 现在有类型安全
+    const claudeClient = new AnthropicAPIClient(providerExtraHeaders)
+    const geminiClient = new GeminiAPIClient({ ...providerExtraHeaders, apiHost: 'https://aihubmix.com/gemini' })
+    const openaiClient = new OpenAIResponseAPIClient(providerExtraHeaders)
+    const defaultClient = new OpenAIAPIClient(providerExtraHeaders)
+
+    this.clients.set('claude', claudeClient)
+    this.clients.set('gemini', geminiClient)
+    this.clients.set('openai', openaiClient)
+    this.clients.set('default', defaultClient)
+
+    // 设置默认client
+    this.defaultClient = defaultClient
+    this.currentClient = this.defaultClient as BaseApiClient
+  }
+
+  override getBaseURL(): string {
+    if (!this.currentClient) {
+      return this.provider.apiHost
+    }
+    return this.currentClient.getBaseURL()
+  }
+
+  /**
+   * 类型守卫：确保client是BaseApiClient的实例
+   */
+  private isValidClient(client: unknown): client is BaseApiClient {
+    return (
+      client !== null &&
+      client !== undefined &&
+      typeof client === 'object' &&
+      'createCompletions' in client &&
+      'getRequestTransformer' in client &&
+      'getResponseChunkTransformer' in client
+    )
+  }
+
+  /**
+   * 根据模型获取合适的client
+   */
+  private getClient(model: Model): BaseApiClient {
+    const id = model.id.toLowerCase()
+
+    // claude开头
+    if (id.startsWith('claude')) {
+      const client = this.clients.get('claude')
+      if (!client || !this.isValidClient(client)) {
+        throw new Error('Claude client not properly initialized')
+      }
+      return client
+    }
+
+    // gemini开头 且不以-nothink、-search结尾
+    if ((id.startsWith('gemini') || id.startsWith('imagen')) && !id.endsWith('-nothink') && !id.endsWith('-search')) {
+      const client = this.clients.get('gemini')
+      if (!client || !this.isValidClient(client)) {
+        throw new Error('Gemini client not properly initialized')
+      }
+      return client
+    }
+
+    // OpenAI系列模型
+    if (isOpenAILLMModel(model)) {
+      const client = this.clients.get('openai')
+      if (!client || !this.isValidClient(client)) {
+        throw new Error('OpenAI client not properly initialized')
+      }
+      return client
+    }
+
+    return this.defaultClient as BaseApiClient
+  }
+
+  /**
+   * 根据模型选择合适的client并委托调用
+   */
+  public getClientForModel(model: Model): BaseApiClient {
+    this.currentClient = this.getClient(model)
+    return this.currentClient
+  }
+
+  // ============ BaseApiClient 抽象方法实现 ============
+
+  async createCompletions(payload: SdkParams, options?: RequestOptions): Promise<SdkRawOutput> {
+    // 尝试从payload中提取模型信息来选择client
+    const modelId = this.extractModelFromPayload(payload)
+    if (modelId) {
+      const modelObj = { id: modelId } as Model
+      const targetClient = this.getClient(modelObj)
+      return targetClient.createCompletions(payload, options)
+    }
+
+    // 如果无法从payload中提取模型，使用当前设置的client
+    return this.currentClient.createCompletions(payload, options)
+  }
+
+  /**
+   * 从SDK payload中提取模型ID
+   */
+  private extractModelFromPayload(payload: SdkParams): string | null {
+    // 不同的SDK可能有不同的字段名
+    if ('model' in payload && typeof payload.model === 'string') {
+      return payload.model
+    }
+    return null
+  }
+
+  async generateImage(params: GenerateImageParams): Promise<string[]> {
+    return this.currentClient.generateImage(params)
+  }
+
+  async getEmbeddingDimensions(model?: Model): Promise<number> {
+    const client = model ? this.getClient(model) : this.currentClient
+    return client.getEmbeddingDimensions(model)
+  }
+
+  async listModels(): Promise<SdkModel[]> {
+    // 可以聚合所有client的模型，或者使用默认client
+    return this.defaultClient.listModels()
+  }
+
+  async getSdkInstance(): Promise<SdkInstance> {
+    return this.currentClient.getSdkInstance()
+  }
+
+  getRequestTransformer(): RequestTransformer<SdkParams, SdkMessageParam> {
+    return this.currentClient.getRequestTransformer()
+  }
+
+  getResponseChunkTransformer(ctx: CompletionsContext): ResponseChunkTransformer<SdkRawChunk> {
+    return this.currentClient.getResponseChunkTransformer(ctx)
+  }
+
+  convertMcpToolsToSdkTools(mcpTools: MCPTool[]): SdkTool[] {
+    return this.currentClient.convertMcpToolsToSdkTools(mcpTools)
+  }
+
+  convertSdkToolCallToMcp(toolCall: SdkToolCall, mcpTools: MCPTool[]): MCPTool | undefined {
+    return this.currentClient.convertSdkToolCallToMcp(toolCall, mcpTools)
+  }
+
+  convertSdkToolCallToMcpToolResponse(toolCall: SdkToolCall, mcpTool: MCPTool): ToolCallResponse {
+    return this.currentClient.convertSdkToolCallToMcpToolResponse(toolCall, mcpTool)
+  }
+
+  buildSdkMessages(
+    currentReqMessages: SdkMessageParam[],
+    output: SdkRawOutput | string,
+    toolResults: SdkMessageParam[],
+    toolCalls?: SdkToolCall[]
+  ): SdkMessageParam[] {
+    return this.currentClient.buildSdkMessages(currentReqMessages, output, toolResults, toolCalls)
+  }
+
+  convertMcpToolResponseToSdkMessageParam(
+    mcpToolResponse: MCPToolResponse,
+    resp: MCPCallToolResponse,
+    model: Model
+  ): SdkMessageParam | undefined {
+    const client = this.getClient(model)
+    return client.convertMcpToolResponseToSdkMessageParam(mcpToolResponse, resp, model)
+  }
+
+  extractMessagesFromSdkPayload(sdkPayload: SdkParams): SdkMessageParam[] {
+    return this.currentClient.extractMessagesFromSdkPayload(sdkPayload)
+  }
+
+  estimateMessageTokens(message: SdkMessageParam): number {
+    return this.currentClient.estimateMessageTokens(message)
+  }
+}

src/renderer/src/aiCore/clients/ApiClientFactory.ts

@@ -0,0 +1,72 @@
+import { Provider } from '@renderer/types'
+
+import { AihubmixAPIClient } from './AihubmixAPIClient'
+import { AnthropicAPIClient } from './anthropic/AnthropicAPIClient'
+import { BaseApiClient } from './BaseApiClient'
+import { GeminiAPIClient } from './gemini/GeminiAPIClient'
+import { VertexAPIClient } from './gemini/VertexAPIClient'
+import { OpenAIAPIClient } from './openai/OpenAIApiClient'
+import { OpenAIResponseAPIClient } from './openai/OpenAIResponseAPIClient'
+import { PPIOAPIClient } from './ppio/PPIOAPIClient'
+
+/**
+ * Factory for creating ApiClient instances based on provider configuration
+ * 根据提供者配置创建ApiClient实例的工厂
+ */
+export class ApiClientFactory {
+  /**
+   * Create an ApiClient instance for the given provider
+   * 为给定的提供者创建ApiClient实例
+   */
+  static create(provider: Provider): BaseApiClient {
+    console.log(`[ApiClientFactory] Creating ApiClient for provider:`, {
+      id: provider.id,
+      type: provider.type
+    })
+
+    let instance: BaseApiClient
+
+    // 首先检查特殊的provider id
+    if (provider.id === 'aihubmix') {
+      console.log(`[ApiClientFactory] Creating AihubmixAPIClient for provider: ${provider.id}`)
+      instance = new AihubmixAPIClient(provider) as BaseApiClient
+      return instance
+    }
+    if (provider.id === 'ppio') {
+      console.log(`[ApiClientFactory] Creating PPIOAPIClient for provider: ${provider.id}`)
+      instance = new PPIOAPIClient(provider) as BaseApiClient
+      return instance
+    }
+
+    // 然后检查标准的provider type
+    switch (provider.type) {
+      case 'openai':
+      case 'azure-openai':
+        console.log(`[ApiClientFactory] Creating OpenAIApiClient for provider: ${provider.id}`)
+        instance = new OpenAIAPIClient(provider) as BaseApiClient
+        break
+      case 'openai-response':
+        instance = new OpenAIResponseAPIClient(provider) as BaseApiClient
+        break
+      case 'gemini':
+        instance = new GeminiAPIClient(provider) as BaseApiClient
+        break
+      case 'vertexai':
+        instance = new VertexAPIClient(provider) as BaseApiClient
+        break
+      case 'anthropic':
+        instance = new AnthropicAPIClient(provider) as BaseApiClient
+        break
+      default:
+        console.log(`[ApiClientFactory] Using default OpenAIApiClient for provider: ${provider.id}`)
+        instance = new OpenAIAPIClient(provider) as BaseApiClient
+        break
+    }
+
+    return instance
+  }
+}
+
+export function isOpenAIProvider(provider: Provider) {
+  return !['anthropic', 'gemini'].includes(provider.type)
+}

src/renderer/src/aiCore/clients/BaseApiClient.ts

@@ -1,40 +1,70 @@
-import Logger from '@renderer/config/logger'
-import { isFunctionCallingModel, isNotSupportTemperatureAndTopP } from '@renderer/config/models'
+import {
+  isFunctionCallingModel,
+  isNotSupportTemperatureAndTopP,
+  isOpenAIModel,
+  isSupportedFlexServiceTier
+} from '@renderer/config/models'
 import { REFERENCE_PROMPT } from '@renderer/config/prompts'
 import { getLMStudioKeepAliveTime } from '@renderer/hooks/useLMStudio'
-import type {
+import { getStoreSetting } from '@renderer/hooks/useSettings'
+import { SettingsState } from '@renderer/store/settings'
+import {
   Assistant,
+  FileTypes,
   GenerateImageParams,
   KnowledgeReference,
   MCPCallToolResponse,
   MCPTool,
   MCPToolResponse,
   Model,
+  OpenAIServiceTier,
   Provider,
-  Suggestion,
+  ToolCallResponse,
   WebSearchProviderResponse,
   WebSearchResponse
 } from '@renderer/types'
-import { ChunkType } from '@renderer/types/chunk'
-import type { Message } from '@renderer/types/newMessage'
-import { delay, isJSON, parseJSON } from '@renderer/utils'
+import { Message } from '@renderer/types/newMessage'
+import {
+  RequestOptions,
+  SdkInstance,
+  SdkMessageParam,
+  SdkModel,
+  SdkParams,
+  SdkRawChunk,
+  SdkRawOutput,
+  SdkTool,
+  SdkToolCall
+} from '@renderer/types/sdk'
+import { isJSON, parseJSON } from '@renderer/utils'
 import { addAbortController, removeAbortController } from '@renderer/utils/abortController'
-import { formatApiHost } from '@renderer/utils/api'
-import { getMainTextContent } from '@renderer/utils/messageUtils/find'
+import { findFileBlocks, getContentWithTools, getMainTextContent } from '@renderer/utils/messageUtils/find'
+import { defaultTimeout } from '@shared/config/constant'
+import Logger from 'electron-log/renderer'
 import { isEmpty } from 'lodash'
-import type OpenAI from 'openai'
 
-import type { CompletionsParams } from '.'
+import { CompletionsContext } from '../middleware/types'
+import { ApiClient, RequestTransformer, ResponseChunkTransformer } from './types'
 
-export default abstract class BaseProvider {
-  // Threshold for determining whether to use system prompt for tools
+/**
+ * Abstract base class for API clients.
+ * Provides common functionality and structure for specific client implementations.
+ */
+export abstract class BaseApiClient<
+  TSdkInstance extends SdkInstance = SdkInstance,
+  TSdkParams extends SdkParams = SdkParams,
+  TRawOutput extends SdkRawOutput = SdkRawOutput,
+  TRawChunk extends SdkRawChunk = SdkRawChunk,
+  TMessageParam extends SdkMessageParam = SdkMessageParam,
+  TToolCall extends SdkToolCall = SdkToolCall,
+  TSdkSpecificTool extends SdkTool = SdkTool
+> implements ApiClient<TSdkInstance, TSdkParams, TRawOutput, TRawChunk, TMessageParam, TToolCall, TSdkSpecificTool>
+{
   private static readonly SYSTEM_PROMPT_THRESHOLD: number = 128
-
-  protected provider: Provider
+  public provider: Provider
   protected host: string
   protected apiKey: string
-
-  protected useSystemPromptForTools: boolean = true
+  protected sdkInstance?: TSdkInstance
+  public useSystemPromptForTools: boolean = true
 
   constructor(provider: Provider) {
     this.provider = provider
@@ -42,31 +72,70 @@ export default abstract class BaseProvider {
     this.apiKey = this.getApiKey()
   }
 
-  abstract completions({ messages, assistant, onChunk, onFilterMessages }: CompletionsParams): Promise<void>
-  abstract translate(
-    content: string,
-    assistant: Assistant,
-    onResponse?: (text: string, isComplete: boolean) => void
-  ): Promise<string>
-  abstract summaries(messages: Message[], assistant: Assistant): Promise<string>
-  abstract summaryForSearch(messages: Message[], assistant: Assistant): Promise<string | null>
-  abstract suggestions(messages: Message[], assistant: Assistant): Promise<Suggestion[]>
-  abstract generateText({ prompt, content }: { prompt: string; content: string }): Promise<string>
-  abstract check(model: Model, stream: boolean): Promise<{ valid: boolean; error: Error | null }>
-  abstract models(): Promise<OpenAI.Models.Model[]>
-  abstract generateImage(params: GenerateImageParams): Promise<string[]>
-  abstract generateImageByChat({ messages, assistant, onChunk, onFilterMessages }: CompletionsParams): Promise<void>
-  abstract getEmbeddingDimensions(model: Model): Promise<number>
-  public abstract convertMcpTools<T>(mcpTools: MCPTool[]): T[]
-  public abstract mcpToolCallResponseToMessage(
+  // // 核心的completions方法 - 在中间件架构中，这通常只是一个占位符
+  // abstract completions(params: CompletionsParams, internal?: ProcessingState): Promise<CompletionsResult>
+
+  /**
+   * 核心API Endpoint
+   **/
+
+  abstract createCompletions(payload: TSdkParams, options?: RequestOptions): Promise<TRawOutput>
+
+  abstract generateImage(generateImageParams: GenerateImageParams): Promise<string[]>
+
+  abstract getEmbeddingDimensions(model?: Model): Promise<number>
+
+  abstract listModels(): Promise<SdkModel[]>
+
+  abstract getSdkInstance(): Promise<TSdkInstance> | TSdkInstance
+
+  /**
+   * 中间件
+   **/
+
+  // 在 CoreRequestToSdkParamsMiddleware中使用
+  abstract getRequestTransformer(): RequestTransformer<TSdkParams, TMessageParam>
+  // 在RawSdkChunkToGenericChunkMiddleware中使用
+  abstract getResponseChunkTransformer(ctx: CompletionsContext): ResponseChunkTransformer<TRawChunk>
+
+  /**
+   * 工具转换
+   **/
+
+  // Optional tool conversion methods - implement if needed by the specific provider
+  abstract convertMcpToolsToSdkTools(mcpTools: MCPTool[]): TSdkSpecificTool[]
+
+  abstract convertSdkToolCallToMcp(toolCall: TToolCall, mcpTools: MCPTool[]): MCPTool | undefined
+
+  abstract convertSdkToolCallToMcpToolResponse(toolCall: TToolCall, mcpTool: MCPTool): ToolCallResponse
+
+  abstract buildSdkMessages(
+    currentReqMessages: TMessageParam[],
+    output: TRawOutput | string | undefined,
+    toolResults: TMessageParam[],
+    toolCalls?: TToolCall[]
+  ): TMessageParam[]
+
+  abstract estimateMessageTokens(message: TMessageParam): number
+
+  abstract convertMcpToolResponseToSdkMessageParam(
     mcpToolResponse: MCPToolResponse,
     resp: MCPCallToolResponse,
     model: Model
-  ): any
+  ): TMessageParam | undefined
+
+  /**
+   * 从SDK载荷中提取消息数组（用于中间件中的类型安全访问）
+   * 不同的提供商可能使用不同的字段名（如messages、history等）
+   */
+  abstract extractMessagesFromSdkPayload(sdkPayload: TSdkParams): TMessageParam[]
+
+  /**
+   * 通用函数
+   **/
 
   public getBaseURL(): string {
-    const host = this.provider.apiHost
-    return formatApiHost(host)
+    return this.provider.apiHost
   }
 
   public getApiKey() {
@@ -111,18 +180,37 @@ export default abstract class BaseProvider {
     return isNotSupportTemperatureAndTopP(model) ? undefined : assistant.settings?.topP
   }
 
-  public async fakeCompletions({ onChunk }: CompletionsParams) {
-    for (let i = 0; i < 100; i++) {
-      await delay(0.01)
-      onChunk({
-        response: { text: i + '\n', usage: { completion_tokens: 0, prompt_tokens: 0, total_tokens: 0 } },
-        type: ChunkType.BLOCK_COMPLETE
-      })
+  protected getServiceTier(model: Model) {
+    if (!isOpenAIModel(model) || model.provider === 'github' || model.provider === 'copilot') {
+      return undefined
     }
+
+    const openAI = getStoreSetting('openAI') as SettingsState['openAI']
+    let serviceTier = 'auto' as OpenAIServiceTier
+
+    if (openAI && openAI?.serviceTier === 'flex') {
+      if (isSupportedFlexServiceTier(model)) {
+        serviceTier = 'flex'
+      } else {
+        serviceTier = 'auto'
+      }
+    } else {
+      serviceTier = openAI.serviceTier
+    }
+
+    return serviceTier
+  }
+
+  protected getTimeout(model: Model) {
+    if (isSupportedFlexServiceTier(model)) {
+      return 15 * 1000 * 60
+    }
+    return defaultTimeout
   }
 
   public async getMessageContent(message: Message): Promise<string> {
-    const content = getMainTextContent(message)
+    const content = getContentWithTools(message)
+
     if (isEmpty(content)) {
       return ''
     }
@@ -148,6 +236,36 @@ export default abstract class BaseProvider {
     return content
   }
 
+  /**
+   * Extract the file content from the message
+   * @param message - The message
+   * @returns The file content
+   */
+  protected async extractFileContent(message: Message) {
+    const fileBlocks = findFileBlocks(message)
+    if (fileBlocks.length > 0) {
+      const textFileBlocks = fileBlocks.filter(
+        (fb) => fb.file && [FileTypes.TEXT, FileTypes.DOCUMENT].includes(fb.file.type)
+      )
+
+      if (textFileBlocks.length > 0) {
+        let text = ''
+        const divider = '\n\n---\n\n'
+
+        for (const fileBlock of textFileBlocks) {
+          const file = fileBlock.file
+          const fileContent = (await window.api.file.read(file.id + file.ext)).trim()
+          const fileNameRow = 'file: ' + file.origin_name + '\n\n'
+          text = text + fileNameRow + fileContent + divider
+        }
+
+        return text
+      }
+    }
+
+    return ''
+  }
+
   private async getWebSearchReferencesFromCache(message: Message) {
     const content = getMainTextContent(message)
     if (isEmpty(content)) {
@@ -156,6 +274,7 @@ export default abstract class BaseProvider {
     const webSearch: WebSearchResponse = window.keyv.get(`web-search-${message.id}`)
 
     if (webSearch) {
+      window.keyv.remove(`web-search-${message.id}`)
       return (webSearch.results as WebSearchProviderResponse).results.map(
         (result, index) =>
           ({
@@ -181,6 +300,7 @@ export default abstract class BaseProvider {
     const knowledgeReferences: KnowledgeReference[] = window.keyv.get(`knowledge-search-${message.id}`)
 
     if (!isEmpty(knowledgeReferences)) {
+      window.keyv.remove(`knowledge-search-${message.id}`)
       // Logger.log(`Found ${knowledgeReferences.length} knowledge base references in cache for ID: ${message.id}`)
       return knowledgeReferences
     }
@@ -209,7 +329,7 @@ export default abstract class BaseProvider {
     )
   }
 
-  protected createAbortController(messageId?: string, isAddEventListener?: boolean) {
+  public createAbortController(messageId?: string, isAddEventListener?: boolean) {
     const abortController = new AbortController()
     const abortFn = () => abortController.abort()
 
@@ -255,11 +375,11 @@ export default abstract class BaseProvider {
   }
 
   // Setup tools configuration based on provided parameters
-  protected setupToolsConfig<T>(params: { mcpTools?: MCPTool[]; model: Model; enableToolUse?: boolean }): {
-    tools: T[]
+  public setupToolsConfig(params: { mcpTools?: MCPTool[]; model: Model; enableToolUse?: boolean }): {
+    tools: TSdkSpecificTool[]
   } {
     const { mcpTools, model, enableToolUse } = params
-    let tools: T[] = []
+    let tools: TSdkSpecificTool[] = []
 
     // If there are no tools, return an empty array
     if (!mcpTools?.length) {
@@ -267,14 +387,14 @@ export default abstract class BaseProvider {
     }
 
     // If the number of tools exceeds the threshold, use the system prompt
-    if (mcpTools.length > BaseProvider.SYSTEM_PROMPT_THRESHOLD) {
+    if (mcpTools.length > BaseApiClient.SYSTEM_PROMPT_THRESHOLD) {
       this.useSystemPromptForTools = true
       return { tools }
     }
 
     // If the model supports function calling and tool usage is enabled
     if (isFunctionCallingModel(model) && enableToolUse) {
-      tools = this.convertMcpTools<T>(mcpTools)
+      tools = this.convertMcpToolsToSdkTools(mcpTools)
       this.useSystemPromptForTools = false
     }
 

src/renderer/src/aiCore/clients/anthropic/AnthropicAPIClient.ts

@@ -0,0 +1,736 @@
+import Anthropic from '@anthropic-ai/sdk'
+import {
+  Base64ImageSource,
+  ImageBlockParam,
+  MessageParam,
+  TextBlockParam,
+  ToolResultBlockParam,
+  ToolUseBlock,
+  WebSearchTool20250305
+} from '@anthropic-ai/sdk/resources'
+import {
+  ContentBlock,
+  ContentBlockParam,
+  MessageCreateParams,
+  MessageCreateParamsBase,
+  RedactedThinkingBlockParam,
+  ServerToolUseBlockParam,
+  ThinkingBlockParam,
+  ThinkingConfigParam,
+  ToolUnion,
+  ToolUseBlockParam,
+  WebSearchResultBlock,
+  WebSearchToolResultBlockParam,
+  WebSearchToolResultError
+} from '@anthropic-ai/sdk/resources/messages'
+import { MessageStream } from '@anthropic-ai/sdk/resources/messages/messages'
+import { GenericChunk } from '@renderer/aiCore/middleware/schemas'
+import { DEFAULT_MAX_TOKENS } from '@renderer/config/constant'
+import Logger from '@renderer/config/logger'
+import { findTokenLimit, isClaudeReasoningModel, isReasoningModel, isWebSearchModel } from '@renderer/config/models'
+import { getAssistantSettings } from '@renderer/services/AssistantService'
+import FileManager from '@renderer/services/FileManager'
+import { estimateTextTokens } from '@renderer/services/TokenService'
+import {
+  Assistant,
+  EFFORT_RATIO,
+  FileTypes,
+  MCPCallToolResponse,
+  MCPTool,
+  MCPToolResponse,
+  Model,
+  Provider,
+  ToolCallResponse,
+  WebSearchSource
+} from '@renderer/types'
+import {
+  ChunkType,
+  ErrorChunk,
+  LLMWebSearchCompleteChunk,
+  LLMWebSearchInProgressChunk,
+  MCPToolCreatedChunk,
+  TextDeltaChunk,
+  ThinkingDeltaChunk
+} from '@renderer/types/chunk'
+import { type Message } from '@renderer/types/newMessage'
+import {
+  AnthropicSdkMessageParam,
+  AnthropicSdkParams,
+  AnthropicSdkRawChunk,
+  AnthropicSdkRawOutput
+} from '@renderer/types/sdk'
+import { addImageFileToContents } from '@renderer/utils/formats'
+import {
+  anthropicToolUseToMcpTool,
+  isEnabledToolUse,
+  mcpToolCallResponseToAnthropicMessage,
+  mcpToolsToAnthropicTools
+} from '@renderer/utils/mcp-tools'
+import { findFileBlocks, findImageBlocks } from '@renderer/utils/messageUtils/find'
+import { buildSystemPrompt } from '@renderer/utils/prompt'
+
+import { BaseApiClient } from '../BaseApiClient'
+import { AnthropicStreamListener, RawStreamListener, RequestTransformer, ResponseChunkTransformer } from '../types'
+
+export class AnthropicAPIClient extends BaseApiClient<
+  Anthropic,
+  AnthropicSdkParams,
+  AnthropicSdkRawOutput,
+  AnthropicSdkRawChunk,
+  AnthropicSdkMessageParam,
+  ToolUseBlock,
+  ToolUnion
+> {
+  constructor(provider: Provider) {
+    super(provider)
+  }
+
+  async getSdkInstance(): Promise<Anthropic> {
+    if (this.sdkInstance) {
+      return this.sdkInstance
+    }
+    this.sdkInstance = new Anthropic({
+      apiKey: this.apiKey,
+      baseURL: this.getBaseURL(),
+      dangerouslyAllowBrowser: true,
+      defaultHeaders: {
+        'anthropic-beta': 'output-128k-2025-02-19',
+        ...this.provider.extra_headers
+      }
+    })
+    return this.sdkInstance
+  }
+
+  override async createCompletions(
+    payload: AnthropicSdkParams,
+    options?: Anthropic.RequestOptions
+  ): Promise<AnthropicSdkRawOutput> {
+    const sdk = await this.getSdkInstance()
+    if (payload.stream) {
+      return sdk.messages.stream(payload, options)
+    }
+    return await sdk.messages.create(payload, options)
+  }
+
+  // @ts-ignore sdk未提供
+  // eslint-disable-next-line @typescript-eslint/no-unused-vars
+  override async generateImage(generateImageParams: GenerateImageParams): Promise<string[]> {
+    return []
+  }
+
+  override async listModels(): Promise<Anthropic.ModelInfo[]> {
+    const sdk = await this.getSdkInstance()
+    const response = await sdk.models.list()
+    return response.data
+  }
+
+  // @ts-ignore sdk未提供
+  override async getEmbeddingDimensions(): Promise<number> {
+    throw new Error("Anthropic SDK doesn't support getEmbeddingDimensions method.")
+  }
+
+  override getTemperature(assistant: Assistant, model: Model): number | undefined {
+    if (assistant.settings?.reasoning_effort && isClaudeReasoningModel(model)) {
+      return undefined
+    }
+    return assistant.settings?.temperature
+  }
+
+  override getTopP(assistant: Assistant, model: Model): number | undefined {
+    if (assistant.settings?.reasoning_effort && isClaudeReasoningModel(model)) {
+      return undefined
+    }
+    return assistant.settings?.topP
+  }
+
+  /**
+   * Get the reasoning effort
+   * @param assistant - The assistant
+   * @param model - The model
+   * @returns The reasoning effort
+   */
+  private getBudgetToken(assistant: Assistant, model: Model): ThinkingConfigParam | undefined {
+    if (!isReasoningModel(model)) {
+      return undefined
+    }
+    const { maxTokens } = getAssistantSettings(assistant)
+
+    const reasoningEffort = assistant?.settings?.reasoning_effort
+
+    if (reasoningEffort === undefined) {
+      return {
+        type: 'disabled'
+      }
+    }
+
+    const effortRatio = EFFORT_RATIO[reasoningEffort]
+
+    const budgetTokens = Math.max(
+      1024,
+      Math.floor(
+        Math.min(
+          (findTokenLimit(model.id)?.max! - findTokenLimit(model.id)?.min!) * effortRatio +
+            findTokenLimit(model.id)?.min!,
+          (maxTokens || DEFAULT_MAX_TOKENS) * effortRatio
+        )
+      )
+    )
+
+    return {
+      type: 'enabled',
+      budget_tokens: budgetTokens
+    }
+  }
+
+  /**
+   * Get the message parameter
+   * @param message - The message
+   * @param model - The model
+   * @returns The message parameter
+   */
+  public async convertMessageToSdkParam(message: Message): Promise<AnthropicSdkMessageParam> {
+    const parts: MessageParam['content'] = [
+      {
+        type: 'text',
+        text: await this.getMessageContent(message)
+      }
+    ]
+
+    // Get and process image blocks
+    const imageBlocks = findImageBlocks(message)
+    for (const imageBlock of imageBlocks) {
+      if (imageBlock.file) {
+        // Handle uploaded file
+        const file = imageBlock.file
+        const base64Data = await window.api.file.base64Image(file.id + file.ext)
+        parts.push({
+          type: 'image',
+          source: {
+            data: base64Data.base64,
+            media_type: base64Data.mime.replace('jpg', 'jpeg') as any,
+            type: 'base64'
+          }
+        })
+      }
+    }
+    // Get and process file blocks
+    const fileBlocks = findFileBlocks(message)
+    for (const fileBlock of fileBlocks) {
+      const { file } = fileBlock
+      if ([FileTypes.TEXT, FileTypes.DOCUMENT].includes(file.type)) {
+        if (file.ext === '.pdf' && file.size < 32 * 1024 * 1024) {
+          const base64Data = await FileManager.readBase64File(file)
+          parts.push({
+            type: 'document',
+            source: {
+              type: 'base64',
+              media_type: 'application/pdf',
+              data: base64Data
+            }
+          })
+        } else {
+          const fileContent = await (await window.api.file.read(file.id + file.ext)).trim()
+          parts.push({
+            type: 'text',
+            text: file.origin_name + '\n' + fileContent
+          })
+        }
+      }
+    }
+
+    return {
+      role: message.role === 'system' ? 'user' : message.role,
+      content: parts
+    }
+  }
+
+  public convertMcpToolsToSdkTools(mcpTools: MCPTool[]): ToolUnion[] {
+    return mcpToolsToAnthropicTools(mcpTools)
+  }
+
+  public convertMcpToolResponseToSdkMessageParam(
+    mcpToolResponse: MCPToolResponse,
+    resp: MCPCallToolResponse,
+    model: Model
+  ): AnthropicSdkMessageParam | undefined {
+    if ('toolUseId' in mcpToolResponse && mcpToolResponse.toolUseId) {
+      return mcpToolCallResponseToAnthropicMessage(mcpToolResponse, resp, model)
+    } else if ('toolCallId' in mcpToolResponse) {
+      return {
+        role: 'user',
+        content: [
+          {
+            type: 'tool_result',
+            tool_use_id: mcpToolResponse.toolCallId!,
+            content: resp.content
+              .map((item) => {
+                if (item.type === 'text') {
+                  return {
+                    type: 'text',
+                    text: item.text || ''
+                  } satisfies TextBlockParam
+                }
+                if (item.type === 'image') {
+                  return {
+                    type: 'image',
+                    source: {
+                      data: item.data || '',
+                      media_type: (item.mimeType || 'image/png') as Base64ImageSource['media_type'],
+                      type: 'base64'
+                    }
+                  } satisfies ImageBlockParam
+                }
+                return
+              })
+              .filter((n) => typeof n !== 'undefined'),
+            is_error: resp.isError
+          } satisfies ToolResultBlockParam
+        ]
+      }
+    }
+    return
+  }
+
+  // Implementing abstract methods from BaseApiClient
+  convertSdkToolCallToMcp(toolCall: ToolUseBlock, mcpTools: MCPTool[]): MCPTool | undefined {
+    // Based on anthropicToolUseToMcpTool logic in AnthropicProvider
+    // This might need adjustment based on how tool calls are specifically handled in the new structure
+    const mcpTool = anthropicToolUseToMcpTool(mcpTools, toolCall)
+    return mcpTool
+  }
+
+  convertSdkToolCallToMcpToolResponse(toolCall: ToolUseBlock, mcpTool: MCPTool): ToolCallResponse {
+    return {
+      id: toolCall.id,
+      toolCallId: toolCall.id,
+      tool: mcpTool,
+      arguments: toolCall.input as Record<string, unknown>,
+      status: 'pending'
+    } as ToolCallResponse
+  }
+
+  override buildSdkMessages(
+    currentReqMessages: AnthropicSdkMessageParam[],
+    output: Anthropic.Message,
+    toolResults: AnthropicSdkMessageParam[]
+  ): AnthropicSdkMessageParam[] {
+    const assistantMessage: AnthropicSdkMessageParam = {
+      role: output.role,
+      content: convertContentBlocksToParams(output.content)
+    }
+
+    const newMessages: AnthropicSdkMessageParam[] = [...currentReqMessages, assistantMessage]
+    if (toolResults && toolResults.length > 0) {
+      newMessages.push(...toolResults)
+    }
+    return newMessages
+  }
+
+  override estimateMessageTokens(message: AnthropicSdkMessageParam): number {
+    if (typeof message.content === 'string') {
+      return estimateTextTokens(message.content)
+    }
+    return message.content
+      .map((content) => {
+        switch (content.type) {
+          case 'text':
+            return estimateTextTokens(content.text)
+          case 'image':
+            if (content.source.type === 'base64') {
+              return estimateTextTokens(content.source.data)
+            } else {
+              return estimateTextTokens(content.source.url)
+            }
+          case 'tool_use':
+            return estimateTextTokens(JSON.stringify(content.input))
+          case 'tool_result':
+            return estimateTextTokens(JSON.stringify(content.content))
+          default:
+            return 0
+        }
+      })
+      .reduce((acc, curr) => acc + curr, 0)
+  }
+
+  public buildAssistantMessage(message: Anthropic.Message): AnthropicSdkMessageParam {
+    const messageParam: AnthropicSdkMessageParam = {
+      role: message.role,
+      content: convertContentBlocksToParams(message.content)
+    }
+    return messageParam
+  }
+
+  public extractMessagesFromSdkPayload(sdkPayload: AnthropicSdkParams): AnthropicSdkMessageParam[] {
+    return sdkPayload.messages || []
+  }
+
+  /**
+   * Anthropic专用的原始流监听器
+   * 处理MessageStream对象的特定事件
+   */
+  attachRawStreamListener(
+    rawOutput: AnthropicSdkRawOutput,
+    listener: RawStreamListener<AnthropicSdkRawChunk>
+  ): AnthropicSdkRawOutput {
+    console.log(`[AnthropicApiClient] 附加流监听器到原始输出`)
+    // 专用的Anthropic事件处理
+    const anthropicListener = listener as AnthropicStreamListener
+    // 检查是否为MessageStream
+    if (rawOutput instanceof MessageStream) {
+      console.log(`[AnthropicApiClient] 检测到 Anthropic MessageStream，附加专用监听器`)
+
+      if (listener.onStart) {
+        listener.onStart()
+      }
+
+      if (listener.onChunk) {
+        rawOutput.on('streamEvent', (event: AnthropicSdkRawChunk) => {
+          listener.onChunk!(event)
+        })
+      }
+
+      if (anthropicListener.onContentBlock) {
+        rawOutput.on('contentBlock', anthropicListener.onContentBlock)
+      }
+
+      if (anthropicListener.onMessage) {
+        rawOutput.on('finalMessage', anthropicListener.onMessage)
+      }
+
+      if (listener.onEnd) {
+        rawOutput.on('end', () => {
+          listener.onEnd!()
+        })
+      }
+
+      if (listener.onError) {
+        rawOutput.on('error', (error: Error) => {
+          listener.onError!(error)
+        })
+      }
+
+      return rawOutput
+    }
+
+    if (anthropicListener.onMessage) {
+      anthropicListener.onMessage(rawOutput)
+    }
+
+    // 对于非MessageStream响应
+    return rawOutput
+  }
+
+  private async getWebSearchParams(model: Model): Promise<WebSearchTool20250305 | undefined> {
+    if (!isWebSearchModel(model)) {
+      return undefined
+    }
+    return {
+      type: 'web_search_20250305',
+      name: 'web_search',
+      max_uses: 5
+    } as WebSearchTool20250305
+  }
+
+  getRequestTransformer(): RequestTransformer<AnthropicSdkParams, AnthropicSdkMessageParam> {
+    return {
+      transform: async (
+        coreRequest,
+        assistant,
+        model,
+        isRecursiveCall,
+        recursiveSdkMessages
+      ): Promise<{
+        payload: AnthropicSdkParams
+        messages: AnthropicSdkMessageParam[]
+        metadata: Record<string, any>
+      }> => {
+        const { messages, mcpTools, maxTokens, streamOutput, enableWebSearch } = coreRequest
+        // 1. 处理系统消息
+        let systemPrompt = assistant.prompt
+
+        // 2. 设置工具
+        const { tools } = this.setupToolsConfig({
+          mcpTools: mcpTools,
+          model,
+          enableToolUse: isEnabledToolUse(assistant)
+        })
+
+        if (this.useSystemPromptForTools) {
+          systemPrompt = await buildSystemPrompt(systemPrompt, mcpTools, assistant)
+        }
+
+        const systemMessage: TextBlockParam | undefined = systemPrompt
+          ? { type: 'text', text: systemPrompt }
+          : undefined
+
+        // 3. 处理用户消息
+        const sdkMessages: AnthropicSdkMessageParam[] = []
+        if (typeof messages === 'string') {
+          sdkMessages.push({ role: 'user', content: messages })
+        } else {
+          const processedMessages = addImageFileToContents(messages)
+          for (const message of processedMessages) {
+            sdkMessages.push(await this.convertMessageToSdkParam(message))
+          }
+        }
+
+        if (enableWebSearch) {
+          const webSearchTool = await this.getWebSearchParams(model)
+          if (webSearchTool) {
+            tools.push(webSearchTool)
+          }
+        }
+
+        const commonParams: MessageCreateParamsBase = {
+          model: model.id,
+          messages:
+            isRecursiveCall && recursiveSdkMessages && recursiveSdkMessages.length > 0
+              ? recursiveSdkMessages
+              : sdkMessages,
+          max_tokens: maxTokens || DEFAULT_MAX_TOKENS,
+          temperature: this.getTemperature(assistant, model),
+          top_p: this.getTopP(assistant, model),
+          system: systemMessage ? [systemMessage] : undefined,
+          thinking: this.getBudgetToken(assistant, model),
+          tools: tools.length > 0 ? tools : undefined,
+          // 只在对话场景下应用自定义参数，避免影响翻译、总结等其他业务逻辑
+          ...(coreRequest.callType === 'chat' ? this.getCustomParameters(assistant) : {})
+        }
+
+        const finalParams: MessageCreateParams = streamOutput
+          ? {
+              ...commonParams,
+              stream: true
+            }
+          : {
+              ...commonParams,
+              stream: false
+            }
+
+        const timeout = this.getTimeout(model)
+        return { payload: finalParams, messages: sdkMessages, metadata: { timeout } }
+      }
+    }
+  }
+
+  getResponseChunkTransformer(): ResponseChunkTransformer<AnthropicSdkRawChunk> {
+    return () => {
+      let accumulatedJson = ''
+      const toolCalls: Record<number, ToolUseBlock> = {}
+
+      return {
+        async transform(rawChunk: AnthropicSdkRawChunk, controller: TransformStreamDefaultController<GenericChunk>) {
+          switch (rawChunk.type) {
+            case 'message': {
+              let i = 0
+              for (const content of rawChunk.content) {
+                switch (content.type) {
+                  case 'text': {
+                    controller.enqueue({
+                      type: ChunkType.TEXT_DELTA,
+                      text: content.text
+                    } as TextDeltaChunk)
+                    break
+                  }
+                  case 'tool_use': {
+                    toolCalls[i] = content
+                    i++
+                    break
+                  }
+                  case 'thinking': {
+                    controller.enqueue({
+                      type: ChunkType.THINKING_DELTA,
+                      text: content.thinking
+                    } as ThinkingDeltaChunk)
+                    break
+                  }
+                  case 'web_search_tool_result': {
+                    controller.enqueue({
+                      type: ChunkType.LLM_WEB_SEARCH_COMPLETE,
+                      llm_web_search: {
+                        results: content.content,
+                        source: WebSearchSource.ANTHROPIC
+                      }
+                    } as LLMWebSearchCompleteChunk)
+                    break
+                  }
+                }
+              }
+              if (i > 0) {
+                controller.enqueue({
+                  type: ChunkType.MCP_TOOL_CREATED,
+                  tool_calls: Object.values(toolCalls)
+                } as MCPToolCreatedChunk)
+              }
+              controller.enqueue({
+                type: ChunkType.LLM_RESPONSE_COMPLETE,
+                response: {
+                  usage: {
+                    prompt_tokens: rawChunk.usage.input_tokens || 0,
+                    completion_tokens: rawChunk.usage.output_tokens || 0,
+                    total_tokens: (rawChunk.usage.input_tokens || 0) + (rawChunk.usage.output_tokens || 0)
+                  }
+                }
+              })
+              break
+            }
+            case 'content_block_start': {
+              const contentBlock = rawChunk.content_block
+              switch (contentBlock.type) {
+                case 'server_tool_use': {
+                  if (contentBlock.name === 'web_search') {
+                    controller.enqueue({
+                      type: ChunkType.LLM_WEB_SEARCH_IN_PROGRESS
+                    } as LLMWebSearchInProgressChunk)
+                  }
+                  break
+                }
+                case 'web_search_tool_result': {
+                  if (
+                    contentBlock.content &&
+                    (contentBlock.content as WebSearchToolResultError).type === 'web_search_tool_result_error'
+                  ) {
+                    controller.enqueue({
+                      type: ChunkType.ERROR,
+                      error: {
+                        code: (contentBlock.content as WebSearchToolResultError).error_code,
+                        message: (contentBlock.content as WebSearchToolResultError).error_code
+                      }
+                    } as ErrorChunk)
+                  } else {
+                    controller.enqueue({
+                      type: ChunkType.LLM_WEB_SEARCH_COMPLETE,
+                      llm_web_search: {
+                        results: contentBlock.content as Array<WebSearchResultBlock>,
+                        source: WebSearchSource.ANTHROPIC
+                      }
+                    } as LLMWebSearchCompleteChunk)
+                  }
+                  break
+                }
+                case 'tool_use': {
+                  toolCalls[rawChunk.index] = contentBlock
+                  break
+                }
+              }
+              break
+            }
+            case 'content_block_delta': {
+              const messageDelta = rawChunk.delta
+              switch (messageDelta.type) {
+                case 'text_delta': {
+                  if (messageDelta.text) {
+                    controller.enqueue({
+                      type: ChunkType.TEXT_DELTA,
+                      text: messageDelta.text
+                    } as TextDeltaChunk)
+                  }
+                  break
+                }
+                case 'thinking_delta': {
+                  if (messageDelta.thinking) {
+                    controller.enqueue({
+                      type: ChunkType.THINKING_DELTA,
+                      text: messageDelta.thinking
+                    } as ThinkingDeltaChunk)
+                  }
+                  break
+                }
+                case 'input_json_delta': {
+                  if (messageDelta.partial_json) {
+                    accumulatedJson += messageDelta.partial_json
+                  }
+                  break
+                }
+              }
+              break
+            }
+            case 'content_block_stop': {
+              const toolCall = toolCalls[rawChunk.index]
+              if (toolCall) {
+                try {
+                  toolCall.input = JSON.parse(accumulatedJson)
+                  Logger.debug(`Tool call id: ${toolCall.id}, accumulated json: ${accumulatedJson}`)
+                  controller.enqueue({
+                    type: ChunkType.MCP_TOOL_CREATED,
+                    tool_calls: [toolCall]
+                  } as MCPToolCreatedChunk)
+                } catch (error) {
+                  Logger.error(`Error parsing tool call input: ${error}`)
+                }
+              }
+              break
+            }
+            case 'message_delta': {
+              controller.enqueue({
+                type: ChunkType.LLM_RESPONSE_COMPLETE,
+                response: {
+                  usage: {
+                    prompt_tokens: rawChunk.usage.input_tokens || 0,
+                    completion_tokens: rawChunk.usage.output_tokens || 0,
+                    total_tokens: (rawChunk.usage.input_tokens || 0) + (rawChunk.usage.output_tokens || 0)
+                  }
+                }
+              })
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+/**
+ * 将 ContentBlock 数组转换为 ContentBlockParam 数组
+ * 去除服务器生成的额外字段，只保留发送给API所需的字段
+ */
+function convertContentBlocksToParams(contentBlocks: ContentBlock[]): ContentBlockParam[] {
+  return contentBlocks.map((block): ContentBlockParam => {
+    switch (block.type) {
+      case 'text':
+        // TextBlock -> TextBlockParam，去除 citations 等服务器字段
+        return {
+          type: 'text',
+          text: block.text
+        } satisfies TextBlockParam
+      case 'tool_use':
+        // ToolUseBlock -> ToolUseBlockParam
+        return {
+          type: 'tool_use',
+          id: block.id,
+          name: block.name,
+          input: block.input
+        } satisfies ToolUseBlockParam
+      case 'thinking':
+        // ThinkingBlock -> ThinkingBlockParam
+        return {
+          type: 'thinking',
+          thinking: block.thinking,
+          signature: block.signature
+        } satisfies ThinkingBlockParam
+      case 'redacted_thinking':
+        // RedactedThinkingBlock -> RedactedThinkingBlockParam
+        return {
+          type: 'redacted_thinking',
+          data: block.data
+        } satisfies RedactedThinkingBlockParam
+      case 'server_tool_use':
+        // ServerToolUseBlock -> ServerToolUseBlockParam
+        return {
+          type: 'server_tool_use',
+          id: block.id,
+          name: block.name,
+          input: block.input
+        } satisfies ServerToolUseBlockParam
+      case 'web_search_tool_result':
+        // WebSearchToolResultBlock -> WebSearchToolResultBlockParam
+        return {
+          type: 'web_search_tool_result',
+          tool_use_id: block.tool_use_id,
+          content: block.content
+        } satisfies WebSearchToolResultBlockParam
+      default:
+        return block as ContentBlockParam
+    }
+  })
+}

src/renderer/src/aiCore/clients/gemini/GeminiAPIClient.ts

@@ -0,0 +1,767 @@
+import {
+  Content,
+  createPartFromUri,
+  File,
+  FunctionCall,
+  GenerateContentConfig,
+  GenerateImagesConfig,
+  GoogleGenAI,
+  HarmBlockThreshold,
+  HarmCategory,
+  Modality,
+  Model as GeminiModel,
+  Part,
+  SafetySetting,
+  SendMessageParameters,
+  ThinkingConfig,
+  Tool
+} from '@google/genai'
+import { nanoid } from '@reduxjs/toolkit'
+import { GenericChunk } from '@renderer/aiCore/middleware/schemas'
+import {
+  findTokenLimit,
+  GEMINI_FLASH_MODEL_REGEX,
+  isGemmaModel,
+  isSupportedThinkingTokenGeminiModel,
+  isVisionModel
+} from '@renderer/config/models'
+import { estimateTextTokens } from '@renderer/services/TokenService'
+import {
+  Assistant,
+  EFFORT_RATIO,
+  FileMetadata,
+  FileTypes,
+  FileUploadResponse,
+  GenerateImageParams,
+  MCPCallToolResponse,
+  MCPTool,
+  MCPToolResponse,
+  Model,
+  Provider,
+  ToolCallResponse,
+  WebSearchSource
+} from '@renderer/types'
+import { ChunkType, LLMWebSearchCompleteChunk } from '@renderer/types/chunk'
+import { Message } from '@renderer/types/newMessage'
+import {
+  GeminiOptions,
+  GeminiSdkMessageParam,
+  GeminiSdkParams,
+  GeminiSdkRawChunk,
+  GeminiSdkRawOutput,
+  GeminiSdkToolCall
+} from '@renderer/types/sdk'
+import {
+  geminiFunctionCallToMcpTool,
+  isEnabledToolUse,
+  mcpToolCallResponseToGeminiMessage,
+  mcpToolsToGeminiTools
+} from '@renderer/utils/mcp-tools'
+import { findFileBlocks, findImageBlocks, getMainTextContent } from '@renderer/utils/messageUtils/find'
+import { buildSystemPrompt } from '@renderer/utils/prompt'
+import { defaultTimeout, MB } from '@shared/config/constant'
+
+import { BaseApiClient } from '../BaseApiClient'
+import { RequestTransformer, ResponseChunkTransformer } from '../types'
+
+export class GeminiAPIClient extends BaseApiClient<
+  GoogleGenAI,
+  GeminiSdkParams,
+  GeminiSdkRawOutput,
+  GeminiSdkRawChunk,
+  GeminiSdkMessageParam,
+  GeminiSdkToolCall,
+  Tool
+> {
+  constructor(provider: Provider) {
+    super(provider)
+  }
+
+  override async createCompletions(payload: GeminiSdkParams, options?: GeminiOptions): Promise<GeminiSdkRawOutput> {
+    const sdk = await this.getSdkInstance()
+    const { model, history, ...rest } = payload
+    const realPayload: Omit<GeminiSdkParams, 'model'> = {
+      ...rest,
+      config: {
+        ...rest.config,
+        abortSignal: options?.signal,
+        httpOptions: {
+          ...rest.config?.httpOptions,
+          timeout: options?.timeout
+        }
+      }
+    } satisfies SendMessageParameters
+
+    const streamOutput = options?.streamOutput
+
+    const chat = sdk.chats.create({
+      model: model,
+      history: history
+    })
+
+    if (streamOutput) {
+      const stream = chat.sendMessageStream(realPayload)
+      return stream
+    } else {
+      const response = await chat.sendMessage(realPayload)
+      return response
+    }
+  }
+
+  override async generateImage(generateImageParams: GenerateImageParams): Promise<string[]> {
+    const sdk = await this.getSdkInstance()
+    try {
+      const { model, prompt, imageSize, batchSize, signal } = generateImageParams
+      const config: GenerateImagesConfig = {
+        numberOfImages: batchSize,
+        aspectRatio: imageSize,
+        abortSignal: signal,
+        httpOptions: {
+          timeout: defaultTimeout
+        }
+      }
+      const response = await sdk.models.generateImages({
+        model: model,
+        prompt,
+        config
+      })
+
+      if (!response.generatedImages || response.generatedImages.length === 0) {
+        return []
+      }
+
+      const images = response.generatedImages
+        .filter((image) => image.image?.imageBytes)
+        .map((image) => {
+          const dataPrefix = `data:${image.image?.mimeType || 'image/png'};base64,`
+          return dataPrefix + image.image?.imageBytes
+        })
+      //  console.log(response?.generatedImages?.[0]?.image?.imageBytes);
+      return images
+    } catch (error) {
+      console.error('[generateImage] error:', error)
+      throw error
+    }
+  }
+
+  override async getEmbeddingDimensions(model: Model): Promise<number> {
+    const sdk = await this.getSdkInstance()
+
+    const data = await sdk.models.embedContent({
+      model: model.id,
+      contents: [{ role: 'user', parts: [{ text: 'hi' }] }]
+    })
+    return data.embeddings?.[0]?.values?.length || 0
+  }
+
+  override async listModels(): Promise<GeminiModel[]> {
+    const sdk = await this.getSdkInstance()
+    const response = await sdk.models.list()
+    const models: GeminiModel[] = []
+    for await (const model of response) {
+      models.push(model)
+    }
+    return models
+  }
+
+  override async getSdkInstance() {
+    if (this.sdkInstance) {
+      return this.sdkInstance
+    }
+
+    this.sdkInstance = new GoogleGenAI({
+      vertexai: false,
+      apiKey: this.apiKey,
+      apiVersion: this.getApiVersion(),
+      httpOptions: {
+        baseUrl: this.getBaseURL(),
+        apiVersion: this.getApiVersion(),
+        headers: {
+          ...this.provider.extra_headers
+        }
+      }
+    })
+
+    return this.sdkInstance
+  }
+
+  protected getApiVersion(): string {
+    if (this.provider.isVertex) {
+      return 'v1'
+    }
+    return 'v1beta'
+  }
+
+  /**
+   * Handle a PDF file
+   * @param file - The file
+   * @returns The part
+   */
+  private async handlePdfFile(file: FileMetadata): Promise<Part> {
+    const smallFileSize = 20 * MB
+    const isSmallFile = file.size < smallFileSize
+
+    if (isSmallFile) {
+      const { data, mimeType } = await this.base64File(file)
+      return {
+        inlineData: {
+          data,
+          mimeType
+        } as Part['inlineData']
+      }
+    }
+
+    // Retrieve file from Gemini uploaded files
+    const fileMetadata: FileUploadResponse = await window.api.fileService.retrieve(this.provider, file.id)
+
+    if (fileMetadata.status === 'success') {
+      const remoteFile = fileMetadata.originalFile?.file as File
+      return createPartFromUri(remoteFile.uri!, remoteFile.mimeType!)
+    }
+
+    // If file is not found, upload it to Gemini
+    const result = await window.api.fileService.upload(this.provider, file)
+    const remoteFile = result.originalFile?.file as File
+    return createPartFromUri(remoteFile.uri!, remoteFile.mimeType!)
+  }
+
+  /**
+   * Get the message contents
+   * @param message - The message
+   * @returns The message contents
+   */
+  private async convertMessageToSdkParam(message: Message): Promise<Content> {
+    const role = message.role === 'user' ? 'user' : 'model'
+    const parts: Part[] = [{ text: await this.getMessageContent(message) }]
+
+    // Add any generated images from previous responses
+    const imageBlocks = findImageBlocks(message)
+    for (const imageBlock of imageBlocks) {
+      if (
+        imageBlock.metadata?.generateImageResponse?.images &&
+        imageBlock.metadata.generateImageResponse.images.length > 0
+      ) {
+        for (const imageUrl of imageBlock.metadata.generateImageResponse.images) {
+          if (imageUrl && imageUrl.startsWith('data:')) {
+            // Extract base64 data and mime type from the data URL
+            const matches = imageUrl.match(/^data:(.+);base64,(.*)$/)
+            if (matches && matches.length === 3) {
+              const mimeType = matches[1]
+              const base64Data = matches[2]
+              parts.push({
+                inlineData: {
+                  data: base64Data,
+                  mimeType: mimeType
+                } as Part['inlineData']
+              })
+            }
+          }
+        }
+      }
+      const file = imageBlock.file
+      if (file) {
+        const base64Data = await window.api.file.base64Image(file.id + file.ext)
+        parts.push({
+          inlineData: {
+            data: base64Data.base64,
+            mimeType: base64Data.mime
+          } as Part['inlineData']
+        })
+      }
+    }
+
+    const fileBlocks = findFileBlocks(message)
+    for (const fileBlock of fileBlocks) {
+      const file = fileBlock.file
+      if (file.type === FileTypes.IMAGE) {
+        const base64Data = await window.api.file.base64Image(file.id + file.ext)
+        parts.push({
+          inlineData: {
+            data: base64Data.base64,
+            mimeType: base64Data.mime
+          } as Part['inlineData']
+        })
+      }
+
+      if (file.ext === '.pdf') {
+        parts.push(await this.handlePdfFile(file))
+        continue
+      }
+      if ([FileTypes.TEXT, FileTypes.DOCUMENT].includes(file.type)) {
+        const fileContent = await (await window.api.file.read(file.id + file.ext)).trim()
+        parts.push({
+          text: file.origin_name + '\n' + fileContent
+        })
+      }
+    }
+
+    return {
+      role,
+      parts: parts
+    }
+  }
+
+  // @ts-ignore unused
+  private async getImageFileContents(message: Message): Promise<Content> {
+    const role = message.role === 'user' ? 'user' : 'model'
+    const content = getMainTextContent(message)
+    const parts: Part[] = [{ text: content }]
+    const imageBlocks = findImageBlocks(message)
+    for (const imageBlock of imageBlocks) {
+      if (
+        imageBlock.metadata?.generateImageResponse?.images &&
+        imageBlock.metadata.generateImageResponse.images.length > 0
+      ) {
+        for (const imageUrl of imageBlock.metadata.generateImageResponse.images) {
+          if (imageUrl && imageUrl.startsWith('data:')) {
+            // Extract base64 data and mime type from the data URL
+            const matches = imageUrl.match(/^data:(.+);base64,(.*)$/)
+            if (matches && matches.length === 3) {
+              const mimeType = matches[1]
+              const base64Data = matches[2]
+              parts.push({
+                inlineData: {
+                  data: base64Data,
+                  mimeType: mimeType
+                } as Part['inlineData']
+              })
+            }
+          }
+        }
+      }
+      const file = imageBlock.file
+      if (file) {
+        const base64Data = await window.api.file.base64Image(file.id + file.ext)
+        parts.push({
+          inlineData: {
+            data: base64Data.base64,
+            mimeType: base64Data.mime
+          } as Part['inlineData']
+        })
+      }
+    }
+    return {
+      role,
+      parts: parts
+    }
+  }
+
+  /**
+   * Get the safety settings
+   * @returns The safety settings
+   */
+  private getSafetySettings(): SafetySetting[] {
+    const safetyThreshold = 'OFF' as HarmBlockThreshold
+
+    return [
+      {
+        category: HarmCategory.HARM_CATEGORY_HATE_SPEECH,
+        threshold: safetyThreshold
+      },
+      {
+        category: HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT,
+        threshold: safetyThreshold
+      },
+      {
+        category: HarmCategory.HARM_CATEGORY_HARASSMENT,
+        threshold: safetyThreshold
+      },
+      {
+        category: HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
+        threshold: safetyThreshold
+      },
+      {
+        category: HarmCategory.HARM_CATEGORY_CIVIC_INTEGRITY,
+        threshold: HarmBlockThreshold.BLOCK_NONE
+      }
+    ]
+  }
+
+  /**
+   * Get the reasoning effort for the assistant
+   * @param assistant - The assistant
+   * @param model - The model
+   * @returns The reasoning effort
+   */
+  private getBudgetToken(assistant: Assistant, model: Model) {
+    if (isSupportedThinkingTokenGeminiModel(model)) {
+      const reasoningEffort = assistant?.settings?.reasoning_effort
+
+      // 如果thinking_budget是undefined，不思考
+      if (reasoningEffort === undefined) {
+        return GEMINI_FLASH_MODEL_REGEX.test(model.id)
+          ? {
+              thinkingConfig: {
+                thinkingBudget: 0
+              }
+            }
+          : {}
+      }
+
+      if (reasoningEffort === 'auto') {
+        return {
+          thinkingConfig: {
+            includeThoughts: true,
+            thinkingBudget: -1
+          }
+        }
+      }
+      const effortRatio = EFFORT_RATIO[reasoningEffort]
+      const { min, max } = findTokenLimit(model.id) || { min: 0, max: 0 }
+      // 计算 budgetTokens，确保不低于 min
+      const budget = Math.floor((max - min) * effortRatio + min)
+
+      return {
+        thinkingConfig: {
+          ...(budget > 0 ? { thinkingBudget: budget } : {}),
+          includeThoughts: true
+        } as ThinkingConfig
+      }
+    }
+
+    return {}
+  }
+
+  private getGenerateImageParameter(): Partial<GenerateContentConfig> {
+    return {
+      systemInstruction: undefined,
+      responseModalities: [Modality.TEXT, Modality.IMAGE],
+      responseMimeType: 'text/plain'
+    }
+  }
+
+  getRequestTransformer(): RequestTransformer<GeminiSdkParams, GeminiSdkMessageParam> {
+    return {
+      transform: async (
+        coreRequest,
+        assistant,
+        model,
+        isRecursiveCall,
+        recursiveSdkMessages
+      ): Promise<{
+        payload: GeminiSdkParams
+        messages: GeminiSdkMessageParam[]
+        metadata: Record<string, any>
+      }> => {
+        const { messages, mcpTools, maxTokens, enableWebSearch, enableGenerateImage } = coreRequest
+        // 1. 处理系统消息
+        let systemInstruction = assistant.prompt
+
+        // 2. 设置工具
+        const { tools } = this.setupToolsConfig({
+          mcpTools,
+          model,
+          enableToolUse: isEnabledToolUse(assistant)
+        })
+
+        if (this.useSystemPromptForTools) {
+          systemInstruction = await buildSystemPrompt(assistant.prompt || '', mcpTools, assistant)
+        }
+
+        let messageContents: Content = { role: 'user', parts: [] } // Initialize messageContents
+        const history: Content[] = []
+        // 3. 处理用户消息
+        if (typeof messages === 'string') {
+          messageContents = {
+            role: 'user',
+            parts: [{ text: messages }]
+          }
+        } else {
+          const userLastMessage = messages.pop()
+          if (userLastMessage) {
+            messageContents = await this.convertMessageToSdkParam(userLastMessage)
+            for (const message of messages) {
+              history.push(await this.convertMessageToSdkParam(message))
+            }
+            messages.push(userLastMessage)
+          }
+        }
+
+        if (enableWebSearch) {
+          tools.push({
+            googleSearch: {}
+          })
+        }
+
+        if (isGemmaModel(model) && assistant.prompt) {
+          const isFirstMessage = history.length === 0
+          if (isFirstMessage && messageContents) {
+            const userMessageText =
+              messageContents.parts && messageContents.parts.length > 0
+                ? (messageContents.parts[0] as Part).text || ''
+                : ''
+            const systemMessage = [
+              {
+                text:
+                  '<start_of_turn>user\n' +
+                  systemInstruction +
+                  '<end_of_turn>\n' +
+                  '<start_of_turn>user\n' +
+                  userMessageText +
+                  '<end_of_turn>'
+              }
+            ] as Part[]
+            if (messageContents && messageContents.parts) {
+              messageContents.parts[0] = systemMessage[0]
+            }
+          }
+        }
+
+        const newHistory =
+          isRecursiveCall && recursiveSdkMessages && recursiveSdkMessages.length > 0
+            ? recursiveSdkMessages.slice(0, recursiveSdkMessages.length - 1)
+            : history
+
+        const newMessageContents =
+          isRecursiveCall && recursiveSdkMessages && recursiveSdkMessages.length > 0
+            ? recursiveSdkMessages[recursiveSdkMessages.length - 1]
+            : messageContents
+
+        const generateContentConfig: GenerateContentConfig = {
+          safetySettings: this.getSafetySettings(),
+          systemInstruction: isGemmaModel(model) ? undefined : systemInstruction,
+          temperature: this.getTemperature(assistant, model),
+          topP: this.getTopP(assistant, model),
+          maxOutputTokens: maxTokens,
+          tools: tools,
+          ...(enableGenerateImage ? this.getGenerateImageParameter() : {}),
+          ...this.getBudgetToken(assistant, model),
+          // 只在对话场景下应用自定义参数，避免影响翻译、总结等其他业务逻辑
+          ...(coreRequest.callType === 'chat' ? this.getCustomParameters(assistant) : {})
+        }
+
+        const param: GeminiSdkParams = {
+          model: model.id,
+          config: generateContentConfig,
+          history: newHistory,
+          message: newMessageContents.parts!
+        }
+
+        return {
+          payload: param,
+          messages: [messageContents],
+          metadata: {}
+        }
+      }
+    }
+  }
+
+  getResponseChunkTransformer(): ResponseChunkTransformer<GeminiSdkRawChunk> {
+    return () => ({
+      async transform(chunk: GeminiSdkRawChunk, controller: TransformStreamDefaultController<GenericChunk>) {
+        const toolCalls: FunctionCall[] = []
+        if (chunk.candidates && chunk.candidates.length > 0) {
+          for (const candidate of chunk.candidates) {
+            if (candidate.content) {
+              candidate.content.parts?.forEach((part) => {
+                const text = part.text || ''
+                if (part.thought) {
+                  controller.enqueue({
+                    type: ChunkType.THINKING_DELTA,
+                    text: text
+                  })
+                } else if (part.text) {
+                  controller.enqueue({
+                    type: ChunkType.TEXT_DELTA,
+                    text: text
+                  })
+                } else if (part.inlineData) {
+                  controller.enqueue({
+                    type: ChunkType.IMAGE_COMPLETE,
+                    image: {
+                      type: 'base64',
+                      images: [
+                        part.inlineData?.data?.startsWith('data:')
+                          ? part.inlineData?.data
+                          : `data:${part.inlineData?.mimeType || 'image/png'};base64,${part.inlineData?.data}`
+                      ]
+                    }
+                  })
+                } else if (part.functionCall) {
+                  toolCalls.push(part.functionCall)
+                }
+              })
+            }
+
+            if (candidate.finishReason) {
+              if (candidate.groundingMetadata) {
+                controller.enqueue({
+                  type: ChunkType.LLM_WEB_SEARCH_COMPLETE,
+                  llm_web_search: {
+                    results: candidate.groundingMetadata,
+                    source: WebSearchSource.GEMINI
+                  }
+                } as LLMWebSearchCompleteChunk)
+              }
+              controller.enqueue({
+                type: ChunkType.LLM_RESPONSE_COMPLETE,
+                response: {
+                  usage: {
+                    prompt_tokens: chunk.usageMetadata?.promptTokenCount || 0,
+                    completion_tokens:
+                      (chunk.usageMetadata?.totalTokenCount || 0) - (chunk.usageMetadata?.promptTokenCount || 0),
+                    total_tokens: chunk.usageMetadata?.totalTokenCount || 0
+                  }
+                }
+              })
+            }
+          }
+        }
+
+        if (toolCalls.length > 0) {
+          controller.enqueue({
+            type: ChunkType.MCP_TOOL_CREATED,
+            tool_calls: toolCalls
+          })
+        }
+      }
+    })
+  }
+
+  public convertMcpToolsToSdkTools(mcpTools: MCPTool[]): Tool[] {
+    return mcpToolsToGeminiTools(mcpTools)
+  }
+
+  public convertSdkToolCallToMcp(toolCall: GeminiSdkToolCall, mcpTools: MCPTool[]): MCPTool | undefined {
+    return geminiFunctionCallToMcpTool(mcpTools, toolCall)
+  }
+
+  public convertSdkToolCallToMcpToolResponse(toolCall: GeminiSdkToolCall, mcpTool: MCPTool): ToolCallResponse {
+    const parsedArgs = (() => {
+      try {
+        return typeof toolCall.args === 'string' ? JSON.parse(toolCall.args) : toolCall.args
+      } catch {
+        return toolCall.args
+      }
+    })()
+
+    return {
+      id: toolCall.id || nanoid(),
+      toolCallId: toolCall.id,
+      tool: mcpTool,
+      arguments: parsedArgs,
+      status: 'pending'
+    } as ToolCallResponse
+  }
+
+  public convertMcpToolResponseToSdkMessageParam(
+    mcpToolResponse: MCPToolResponse,
+    resp: MCPCallToolResponse,
+    model: Model
+  ): GeminiSdkMessageParam | undefined {
+    if ('toolUseId' in mcpToolResponse && mcpToolResponse.toolUseId) {
+      return mcpToolCallResponseToGeminiMessage(mcpToolResponse, resp, isVisionModel(model))
+    } else if ('toolCallId' in mcpToolResponse) {
+      return {
+        role: 'user',
+        parts: [
+          {
+            functionResponse: {
+              id: mcpToolResponse.toolCallId,
+              name: mcpToolResponse.tool.id,
+              response: {
+                output: !resp.isError ? resp.content : undefined,
+                error: resp.isError ? resp.content : undefined
+              }
+            }
+          }
+        ]
+      } satisfies Content
+    }
+    return
+  }
+
+  public buildSdkMessages(
+    currentReqMessages: Content[],
+    output: string,
+    toolResults: Content[],
+    toolCalls: FunctionCall[]
+  ): Content[] {
+    const parts: Part[] = []
+    const modelParts: Part[] = []
+    if (output) {
+      modelParts.push({
+        text: output
+      })
+    }
+
+    toolCalls.forEach((toolCall) => {
+      modelParts.push({
+        functionCall: toolCall
+      })
+    })
+
+    parts.push(
+      ...toolResults
+        .map((ts) => ts.parts)
+        .flat()
+        .filter((p) => p !== undefined)
+    )
+
+    const userMessage: Content = {
+      role: 'user',
+      parts: []
+    }
+
+    if (modelParts.length > 0) {
+      currentReqMessages.push({
+        role: 'model',
+        parts: modelParts
+      })
+    }
+    if (parts.length > 0) {
+      userMessage.parts?.push(...parts)
+      currentReqMessages.push(userMessage)
+    }
+
+    return currentReqMessages
+  }
+
+  override estimateMessageTokens(message: GeminiSdkMessageParam): number {
+    return (
+      message.parts?.reduce((acc, part) => {
+        if (part.text) {
+          return acc + estimateTextTokens(part.text)
+        }
+        if (part.functionCall) {
+          return acc + estimateTextTokens(JSON.stringify(part.functionCall))
+        }
+        if (part.functionResponse) {
+          return acc + estimateTextTokens(JSON.stringify(part.functionResponse.response))
+        }
+        if (part.inlineData) {
+          return acc + estimateTextTokens(part.inlineData.data || '')
+        }
+        if (part.fileData) {
+          return acc + estimateTextTokens(part.fileData.fileUri || '')
+        }
+        return acc
+      }, 0) || 0
+    )
+  }
+
+  public extractMessagesFromSdkPayload(sdkPayload: GeminiSdkParams): GeminiSdkMessageParam[] {
+    const messageParam: GeminiSdkMessageParam = {
+      role: 'user',
+      parts: []
+    }
+    if (Array.isArray(sdkPayload.message)) {
+      sdkPayload.message.forEach((part) => {
+        if (typeof part === 'string') {
+          messageParam.parts?.push({ text: part })
+        } else if (typeof part === 'object') {
+          messageParam.parts?.push(part)
+        }
+      })
+    }
+    return [...(sdkPayload.history || []), messageParam]
+  }
+
+  private async base64File(file: FileMetadata) {
+    const { data } = await window.api.file.base64File(file.id + file.ext)
+    return {
+      data,
+      mimeType: 'application/pdf'
+    }
+  }
+}

src/renderer/src/aiCore/clients/gemini/VertexAPIClient.ts

@@ -0,0 +1,95 @@
+import { GoogleGenAI } from '@google/genai'
+import { getVertexAILocation, getVertexAIProjectId, getVertexAIServiceAccount } from '@renderer/hooks/useVertexAI'
+import { Provider } from '@renderer/types'
+
+import { GeminiAPIClient } from './GeminiAPIClient'
+
+export class VertexAPIClient extends GeminiAPIClient {
+  private authHeaders?: Record<string, string>
+  private authHeadersExpiry?: number
+
+  constructor(provider: Provider) {
+    super(provider)
+  }
+
+  override async getSdkInstance() {
+    if (this.sdkInstance) {
+      return this.sdkInstance
+    }
+
+    const serviceAccount = getVertexAIServiceAccount()
+    const projectId = getVertexAIProjectId()
+    const location = getVertexAILocation()
+
+    if (!serviceAccount.privateKey || !serviceAccount.clientEmail || !projectId || !location) {
+      throw new Error('Vertex AI settings are not configured')
+    }
+
+    const authHeaders = await this.getServiceAccountAuthHeaders()
+
+    this.sdkInstance = new GoogleGenAI({
+      vertexai: true,
+      project: projectId,
+      location: location,
+      httpOptions: {
+        apiVersion: this.getApiVersion(),
+        headers: authHeaders
+      }
+    })
+
+    return this.sdkInstance
+  }
+
+  /**
+   * 获取认证头，如果配置了 service account 则从主进程获取
+   */
+  private async getServiceAccountAuthHeaders(): Promise<Record<string, string> | undefined> {
+    const serviceAccount = getVertexAIServiceAccount()
+    const projectId = getVertexAIProjectId()
+
+    // 检查是否配置了 service account
+    if (!serviceAccount.privateKey || !serviceAccount.clientEmail || !projectId) {
+      return undefined
+    }
+
+    // 检查是否已有有效的认证头（提前 5 分钟过期）
+    const now = Date.now()
+    if (this.authHeaders && this.authHeadersExpiry && this.authHeadersExpiry - now > 5 * 60 * 1000) {
+      return this.authHeaders
+    }
+
+    try {
+      // 从主进程获取认证头
+      this.authHeaders = await window.api.vertexAI.getAuthHeaders({
+        projectId,
+        serviceAccount: {
+          privateKey: serviceAccount.privateKey,
+          clientEmail: serviceAccount.clientEmail
+        }
+      })
+
+      // 设置过期时间（通常认证头有效期为 1 小时）
+      this.authHeadersExpiry = now + 60 * 60 * 1000
+
+      return this.authHeaders
+    } catch (error: any) {
+      console.error('Failed to get auth headers:', error)
+      throw new Error(`Service Account authentication failed: ${error.message}`)
+    }
+  }
+
+  /**
+   * 清理认证缓存并重新初始化
+   */
+  clearAuthCache(): void {
+    this.authHeaders = undefined
+    this.authHeadersExpiry = undefined
+
+    const serviceAccount = getVertexAIServiceAccount()
+    const projectId = getVertexAIProjectId()
+
+    if (projectId && serviceAccount.clientEmail) {
+      window.api.vertexAI.clearAuthCache(projectId, serviceAccount.clientEmail)
+    }
+  }
+}

src/renderer/src/aiCore/clients/index.ts

@@ -0,0 +1,6 @@
+export * from './ApiClientFactory'
+export * from './BaseApiClient'
+export * from './types'
+
+// Export specific clients from subdirectories
+export * from './openai/OpenAIApiClient'

src/renderer/src/aiCore/clients/openai/OpenAIApiClient.ts

@@ -0,0 +1,764 @@
+import { DEFAULT_MAX_TOKENS } from '@renderer/config/constant'
+import Logger from '@renderer/config/logger'
+import {
+  findTokenLimit,
+  GEMINI_FLASH_MODEL_REGEX,
+  getOpenAIWebSearchParams,
+  isDoubaoThinkingAutoModel,
+  isReasoningModel,
+  isSupportedReasoningEffortGrokModel,
+  isSupportedReasoningEffortModel,
+  isSupportedReasoningEffortOpenAIModel,
+  isSupportedThinkingTokenClaudeModel,
+  isSupportedThinkingTokenDoubaoModel,
+  isSupportedThinkingTokenGeminiModel,
+  isSupportedThinkingTokenModel,
+  isSupportedThinkingTokenQwenModel,
+  isVisionModel
+} from '@renderer/config/models'
+import { processPostsuffixQwen3Model, processReqMessages } from '@renderer/services/ModelMessageService'
+import { estimateTextTokens } from '@renderer/services/TokenService'
+// For Copilot token
+import {
+  Assistant,
+  EFFORT_RATIO,
+  FileTypes,
+  MCPCallToolResponse,
+  MCPTool,
+  MCPToolResponse,
+  Model,
+  Provider,
+  ToolCallResponse,
+  WebSearchSource
+} from '@renderer/types'
+import { ChunkType } from '@renderer/types/chunk'
+import { Message } from '@renderer/types/newMessage'
+import {
+  OpenAISdkMessageParam,
+  OpenAISdkParams,
+  OpenAISdkRawChunk,
+  OpenAISdkRawContentSource,
+  OpenAISdkRawOutput,
+  ReasoningEffortOptionalParams
+} from '@renderer/types/sdk'
+import { addImageFileToContents } from '@renderer/utils/formats'
+import {
+  isEnabledToolUse,
+  mcpToolCallResponseToOpenAICompatibleMessage,
+  mcpToolsToOpenAIChatTools,
+  openAIToolsToMcpTool
+} from '@renderer/utils/mcp-tools'
+import { findFileBlocks, findImageBlocks } from '@renderer/utils/messageUtils/find'
+import { buildSystemPrompt } from '@renderer/utils/prompt'
+import OpenAI, { AzureOpenAI } from 'openai'
+import { ChatCompletionContentPart, ChatCompletionContentPartRefusal, ChatCompletionTool } from 'openai/resources'
+
+import { GenericChunk } from '../../middleware/schemas'
+import { RequestTransformer, ResponseChunkTransformer, ResponseChunkTransformerContext } from '../types'
+import { OpenAIBaseClient } from './OpenAIBaseClient'
+
+export class OpenAIAPIClient extends OpenAIBaseClient<
+  OpenAI | AzureOpenAI,
+  OpenAISdkParams,
+  OpenAISdkRawOutput,
+  OpenAISdkRawChunk,
+  OpenAISdkMessageParam,
+  OpenAI.Chat.Completions.ChatCompletionMessageToolCall,
+  ChatCompletionTool
+> {
+  constructor(provider: Provider) {
+    super(provider)
+  }
+
+  override async createCompletions(
+    payload: OpenAISdkParams,
+    options?: OpenAI.RequestOptions
+  ): Promise<OpenAISdkRawOutput> {
+    const sdk = await this.getSdkInstance()
+    // @ts-ignore - SDK参数可能有额外的字段
+    return await sdk.chat.completions.create(payload, options)
+  }
+
+  /**
+   * Get the reasoning effort for the assistant
+   * @param assistant - The assistant
+   * @param model - The model
+   * @returns The reasoning effort
+   */
+  // Method for reasoning effort, moved from OpenAIProvider
+  override getReasoningEffort(assistant: Assistant, model: Model): ReasoningEffortOptionalParams {
+    if (this.provider.id === 'groq') {
+      return {}
+    }
+
+    if (!isReasoningModel(model)) {
+      return {}
+    }
+    const reasoningEffort = assistant?.settings?.reasoning_effort
+
+    // Doubao 思考模式支持
+    if (isSupportedThinkingTokenDoubaoModel(model)) {
+      // reasoningEffort 为空，默认开启 enabled
+      if (!reasoningEffort) {
+        return { thinking: { type: 'disabled' } }
+      }
+      if (reasoningEffort === 'high') {
+        return { thinking: { type: 'enabled' } }
+      }
+      if (reasoningEffort === 'auto' && isDoubaoThinkingAutoModel(model)) {
+        return { thinking: { type: 'auto' } }
+      }
+      // 其他情况不带 thinking 字段
+      return {}
+    }
+
+    if (!reasoningEffort) {
+      if (model.provider === 'openrouter') {
+        if (isSupportedThinkingTokenGeminiModel(model) && !GEMINI_FLASH_MODEL_REGEX.test(model.id)) {
+          return {}
+        }
+        return { reasoning: { enabled: false, exclude: true } }
+      }
+      if (isSupportedThinkingTokenQwenModel(model)) {
+        return { enable_thinking: false }
+      }
+
+      if (isSupportedThinkingTokenClaudeModel(model)) {
+        return {}
+      }
+
+      if (isSupportedThinkingTokenGeminiModel(model)) {
+        if (GEMINI_FLASH_MODEL_REGEX.test(model.id)) {
+          return {
+            extra_body: {
+              google: {
+                thinking_config: {
+                  thinking_budget: 0
+                }
+              }
+            }
+          }
+        }
+        return {}
+      }
+
+      if (isSupportedThinkingTokenDoubaoModel(model)) {
+        return { thinking: { type: 'disabled' } }
+      }
+
+      return {}
+    }
+    const effortRatio = EFFORT_RATIO[reasoningEffort]
+    const budgetTokens = Math.floor(
+      (findTokenLimit(model.id)?.max! - findTokenLimit(model.id)?.min!) * effortRatio + findTokenLimit(model.id)?.min!
+    )
+
+    // OpenRouter models
+    if (model.provider === 'openrouter') {
+      if (isSupportedReasoningEffortModel(model) || isSupportedThinkingTokenModel(model)) {
+        return {
+          reasoning: {
+            effort: reasoningEffort === 'auto' ? 'medium' : reasoningEffort
+          }
+        }
+      }
+    }
+
+    // Qwen models
+    if (isSupportedThinkingTokenQwenModel(model)) {
+      return {
+        enable_thinking: true,
+        thinking_budget: budgetTokens
+      }
+    }
+
+    // Grok models
+    if (isSupportedReasoningEffortGrokModel(model)) {
+      return {
+        reasoning_effort: reasoningEffort
+      }
+    }
+
+    // OpenAI models
+    if (isSupportedReasoningEffortOpenAIModel(model)) {
+      return {
+        reasoning_effort: reasoningEffort
+      }
+    }
+
+    if (isSupportedThinkingTokenGeminiModel(model)) {
+      if (reasoningEffort === 'auto') {
+        return {
+          extra_body: {
+            google: {
+              thinking_config: {
+                thinking_budget: -1,
+                include_thoughts: true
+              }
+            }
+          }
+        }
+      }
+      return {
+        extra_body: {
+          google: {
+            thinking_config: {
+              thinking_budget: budgetTokens,
+              include_thoughts: true
+            }
+          }
+        }
+      }
+    }
+
+    // Claude models
+    if (isSupportedThinkingTokenClaudeModel(model)) {
+      const maxTokens = assistant.settings?.maxTokens
+      return {
+        thinking: {
+          type: 'enabled',
+          budget_tokens: Math.floor(
+            Math.max(1024, Math.min(budgetTokens, (maxTokens || DEFAULT_MAX_TOKENS) * effortRatio))
+          )
+        }
+      }
+    }
+
+    // Doubao models
+    if (isSupportedThinkingTokenDoubaoModel(model)) {
+      if (assistant.settings?.reasoning_effort === 'high') {
+        return {
+          thinking: {
+            type: 'enabled'
+          }
+        }
+      }
+    }
+
+    // Default case: no special thinking settings
+    return {}
+  }
+
+  /**
+   * Check if the provider does not support files
+   * @returns True if the provider does not support files, false otherwise
+   */
+  private get isNotSupportFiles() {
+    if (this.provider?.isNotSupportArrayContent) {
+      return true
+    }
+
+    const providers = ['deepseek', 'baichuan', 'minimax', 'xirang']
+
+    return providers.includes(this.provider.id)
+  }
+
+  /**
+   * Get the message parameter
+   * @param message - The message
+   * @param model - The model
+   * @returns The message parameter
+   */
+  public async convertMessageToSdkParam(message: Message, model: Model): Promise<OpenAISdkMessageParam> {
+    const isVision = isVisionModel(model)
+    const content = await this.getMessageContent(message)
+    const fileBlocks = findFileBlocks(message)
+    const imageBlocks = findImageBlocks(message)
+
+    if (fileBlocks.length === 0 && imageBlocks.length === 0) {
+      return {
+        role: message.role === 'system' ? 'user' : message.role,
+        content
+      } as OpenAISdkMessageParam
+    }
+
+    // If the model does not support files, extract the file content
+    if (this.isNotSupportFiles) {
+      const fileContent = await this.extractFileContent(message)
+
+      return {
+        role: message.role === 'system' ? 'user' : message.role,
+        content: content + '\n\n---\n\n' + fileContent
+      } as OpenAISdkMessageParam
+    }
+
+    // If the model supports files, add the file content to the message
+    const parts: ChatCompletionContentPart[] = []
+
+    if (content) {
+      parts.push({ type: 'text', text: content })
+    }
+
+    for (const imageBlock of imageBlocks) {
+      if (isVision) {
+        if (imageBlock.file) {
+          const image = await window.api.file.base64Image(imageBlock.file.id + imageBlock.file.ext)
+          parts.push({ type: 'image_url', image_url: { url: image.data } })
+        } else if (imageBlock.url && imageBlock.url.startsWith('data:')) {
+          parts.push({ type: 'image_url', image_url: { url: imageBlock.url } })
+        }
+      }
+    }
+
+    for (const fileBlock of fileBlocks) {
+      const file = fileBlock.file
+      if (!file) {
+        continue
+      }
+
+      if ([FileTypes.TEXT, FileTypes.DOCUMENT].includes(file.type)) {
+        const fileContent = await (await window.api.file.read(file.id + file.ext)).trim()
+        parts.push({
+          type: 'text',
+          text: file.origin_name + '\n' + fileContent
+        })
+      }
+    }
+
+    return {
+      role: message.role === 'system' ? 'user' : message.role,
+      content: parts
+    } as OpenAISdkMessageParam
+  }
+
+  public convertMcpToolsToSdkTools(mcpTools: MCPTool[]): ChatCompletionTool[] {
+    return mcpToolsToOpenAIChatTools(mcpTools)
+  }
+
+  public convertSdkToolCallToMcp(
+    toolCall: OpenAI.Chat.Completions.ChatCompletionMessageToolCall,
+    mcpTools: MCPTool[]
+  ): MCPTool | undefined {
+    return openAIToolsToMcpTool(mcpTools, toolCall)
+  }
+
+  public convertSdkToolCallToMcpToolResponse(
+    toolCall: OpenAI.Chat.Completions.ChatCompletionMessageToolCall,
+    mcpTool: MCPTool
+  ): ToolCallResponse {
+    let parsedArgs: any
+    try {
+      parsedArgs = JSON.parse(toolCall.function.arguments)
+    } catch {
+      parsedArgs = toolCall.function.arguments
+    }
+    return {
+      id: toolCall.id,
+      toolCallId: toolCall.id,
+      tool: mcpTool,
+      arguments: parsedArgs,
+      status: 'pending'
+    } as ToolCallResponse
+  }
+
+  public convertMcpToolResponseToSdkMessageParam(
+    mcpToolResponse: MCPToolResponse,
+    resp: MCPCallToolResponse,
+    model: Model
+  ): OpenAISdkMessageParam | undefined {
+    if ('toolUseId' in mcpToolResponse && mcpToolResponse.toolUseId) {
+      // This case is for Anthropic/Claude like tool usage, OpenAI uses tool_call_id
+      // For OpenAI, we primarily expect toolCallId. This might need adjustment if mixing provider concepts.
+      return mcpToolCallResponseToOpenAICompatibleMessage(mcpToolResponse, resp, isVisionModel(model))
+    } else if ('toolCallId' in mcpToolResponse && mcpToolResponse.toolCallId) {
+      return {
+        role: 'tool',
+        tool_call_id: mcpToolResponse.toolCallId,
+        content: JSON.stringify(resp.content)
+      } as OpenAI.Chat.Completions.ChatCompletionToolMessageParam
+    }
+    return undefined
+  }
+
+  public buildSdkMessages(
+    currentReqMessages: OpenAISdkMessageParam[],
+    output: string | undefined,
+    toolResults: OpenAISdkMessageParam[],
+    toolCalls: OpenAI.Chat.Completions.ChatCompletionMessageToolCall[]
+  ): OpenAISdkMessageParam[] {
+    if (!output && toolCalls.length === 0) {
+      return [...currentReqMessages, ...toolResults]
+    }
+
+    const assistantMessage: OpenAISdkMessageParam = {
+      role: 'assistant',
+      content: output,
+      tool_calls: toolCalls.length > 0 ? toolCalls : undefined
+    }
+    const newReqMessages = [...currentReqMessages, assistantMessage, ...toolResults]
+    return newReqMessages
+  }
+
+  override estimateMessageTokens(message: OpenAISdkMessageParam): number {
+    let sum = 0
+    if (typeof message.content === 'string') {
+      sum += estimateTextTokens(message.content)
+    } else if (Array.isArray(message.content)) {
+      sum += (message.content || [])
+        .map((part: ChatCompletionContentPart | ChatCompletionContentPartRefusal) => {
+          switch (part.type) {
+            case 'text':
+              return estimateTextTokens(part.text)
+            case 'image_url':
+              return estimateTextTokens(part.image_url.url)
+            case 'input_audio':
+              return estimateTextTokens(part.input_audio.data)
+            case 'file':
+              return estimateTextTokens(part.file.file_data || '')
+            default:
+              return 0
+          }
+        })
+        .reduce((acc, curr) => acc + curr, 0)
+    }
+    if ('tool_calls' in message && message.tool_calls) {
+      sum += message.tool_calls.reduce((acc, toolCall) => {
+        return acc + estimateTextTokens(JSON.stringify(toolCall.function.arguments))
+      }, 0)
+    }
+    return sum
+  }
+
+  public extractMessagesFromSdkPayload(sdkPayload: OpenAISdkParams): OpenAISdkMessageParam[] {
+    return sdkPayload.messages || []
+  }
+
+  getRequestTransformer(): RequestTransformer<OpenAISdkParams, OpenAISdkMessageParam> {
+    return {
+      transform: async (
+        coreRequest,
+        assistant,
+        model,
+        isRecursiveCall,
+        recursiveSdkMessages
+      ): Promise<{
+        payload: OpenAISdkParams
+        messages: OpenAISdkMessageParam[]
+        metadata: Record<string, any>
+      }> => {
+        const { messages, mcpTools, maxTokens, streamOutput, enableWebSearch } = coreRequest
+        // 1. 处理系统消息
+        let systemMessage = { role: 'system', content: assistant.prompt || '' }
+
+        if (isSupportedReasoningEffortOpenAIModel(model)) {
+          systemMessage = {
+            role: 'developer',
+            content: `Formatting re-enabled${systemMessage ? '\n' + systemMessage.content : ''}`
+          }
+        }
+
+        if (model.id.includes('o1-mini') || model.id.includes('o1-preview')) {
+          systemMessage.role = 'assistant'
+        }
+
+        // 2. 设置工具（必须在this.usesystemPromptForTools前面）
+        const { tools } = this.setupToolsConfig({
+          mcpTools: mcpTools,
+          model,
+          enableToolUse: isEnabledToolUse(assistant)
+        })
+
+        if (this.useSystemPromptForTools) {
+          systemMessage.content = await buildSystemPrompt(systemMessage.content || '', mcpTools, assistant)
+        }
+
+        // 3. 处理用户消息
+        const userMessages: OpenAISdkMessageParam[] = []
+        if (typeof messages === 'string') {
+          userMessages.push({ role: 'user', content: messages })
+        } else {
+          const processedMessages = addImageFileToContents(messages)
+          for (const message of processedMessages) {
+            userMessages.push(await this.convertMessageToSdkParam(message, model))
+          }
+        }
+
+        const lastUserMsg = userMessages.findLast((m) => m.role === 'user')
+        if (lastUserMsg && isSupportedThinkingTokenQwenModel(model)) {
+          const postsuffix = '/no_think'
+          const qwenThinkModeEnabled = assistant.settings?.qwenThinkMode === true
+          const currentContent = lastUserMsg.content
+
+          lastUserMsg.content = processPostsuffixQwen3Model(currentContent, postsuffix, qwenThinkModeEnabled) as any
+        }
+
+        // 4. 最终请求消息
+        let reqMessages: OpenAISdkMessageParam[]
+        if (!systemMessage.content) {
+          reqMessages = [...userMessages]
+        } else {
+          reqMessages = [systemMessage, ...userMessages].filter(Boolean) as OpenAISdkMessageParam[]
+        }
+
+        reqMessages = processReqMessages(model, reqMessages)
+
+        // 5. 创建通用参数
+        const commonParams = {
+          model: model.id,
+          messages:
+            isRecursiveCall && recursiveSdkMessages && recursiveSdkMessages.length > 0
+              ? recursiveSdkMessages
+              : reqMessages,
+          temperature: this.getTemperature(assistant, model),
+          top_p: this.getTopP(assistant, model),
+          max_tokens: maxTokens,
+          tools: tools.length > 0 ? tools : undefined,
+          service_tier: this.getServiceTier(model),
+          ...this.getProviderSpecificParameters(assistant, model),
+          ...this.getReasoningEffort(assistant, model),
+          ...getOpenAIWebSearchParams(model, enableWebSearch),
+          // 只在对话场景下应用自定义参数，避免影响翻译、总结等其他业务逻辑
+          ...(coreRequest.callType === 'chat' ? this.getCustomParameters(assistant) : {})
+        }
+
+        // Create the appropriate parameters object based on whether streaming is enabled
+        const sdkParams: OpenAISdkParams = streamOutput
+          ? {
+              ...commonParams,
+              stream: true
+            }
+          : {
+              ...commonParams,
+              stream: false
+            }
+
+        const timeout = this.getTimeout(model)
+
+        return { payload: sdkParams, messages: reqMessages, metadata: { timeout } }
+      }
+    }
+  }
+
+  // 在RawSdkChunkToGenericChunkMiddleware中使用
+  getResponseChunkTransformer(): ResponseChunkTransformer<OpenAISdkRawChunk> {
+    let hasBeenCollectedWebSearch = false
+    const collectWebSearchData = (
+      chunk: OpenAISdkRawChunk,
+      contentSource: OpenAISdkRawContentSource,
+      context: ResponseChunkTransformerContext
+    ) => {
+      if (hasBeenCollectedWebSearch) {
+        return
+      }
+      // OpenAI annotations
+      // @ts-ignore - annotations may not be in standard type definitions
+      const annotations = contentSource.annotations || chunk.annotations
+      if (annotations && annotations.length > 0 && annotations[0].type === 'url_citation') {
+        hasBeenCollectedWebSearch = true
+        return {
+          results: annotations,
+          source: WebSearchSource.OPENAI
+        }
+      }
+
+      // Grok citations
+      // @ts-ignore - citations may not be in standard type definitions
+      if (context.provider?.id === 'grok' && chunk.citations) {
+        hasBeenCollectedWebSearch = true
+        return {
+          // @ts-ignore - citations may not be in standard type definitions
+          results: chunk.citations,
+          source: WebSearchSource.GROK
+        }
+      }
+
+      // Perplexity citations
+      // @ts-ignore - citations may not be in standard type definitions
+      if (context.provider?.id === 'perplexity' && chunk.citations && chunk.citations.length > 0) {
+        hasBeenCollectedWebSearch = true
+        return {
+          // @ts-ignore - citations may not be in standard type definitions
+          results: chunk.citations,
+          source: WebSearchSource.PERPLEXITY
+        }
+      }
+
+      // OpenRouter citations
+      // @ts-ignore - citations may not be in standard type definitions
+      if (context.provider?.id === 'openrouter' && chunk.citations && chunk.citations.length > 0) {
+        hasBeenCollectedWebSearch = true
+        return {
+          // @ts-ignore - citations may not be in standard type definitions
+          results: chunk.citations,
+          source: WebSearchSource.OPENROUTER
+        }
+      }
+
+      // Zhipu web search
+      // @ts-ignore - web_search may not be in standard type definitions
+      if (context.provider?.id === 'zhipu' && chunk.web_search) {
+        hasBeenCollectedWebSearch = true
+        return {
+          // @ts-ignore - web_search may not be in standard type definitions
+          results: chunk.web_search,
+          source: WebSearchSource.ZHIPU
+        }
+      }
+
+      // Hunyuan web search
+      // @ts-ignore - search_info may not be in standard type definitions
+      if (context.provider?.id === 'hunyuan' && chunk.search_info?.search_results) {
+        hasBeenCollectedWebSearch = true
+        return {
+          // @ts-ignore - search_info may not be in standard type definitions
+          results: chunk.search_info.search_results,
+          source: WebSearchSource.HUNYUAN
+        }
+      }
+
+      // TODO: 放到AnthropicApiClient中
+      // // Other providers...
+      // // @ts-ignore - web_search may not be in standard type definitions
+      // if (chunk.web_search) {
+      //   const sourceMap: Record<string, string> = {
+      //     openai: 'openai',
+      //     anthropic: 'anthropic',
+      //     qwenlm: 'qwen'
+      //   }
+      //   const source = sourceMap[context.provider?.id] || 'openai_response'
+      //   return {
+      //     results: chunk.web_search,
+      //     source: source as const
+      //   }
+      // }
+
+      return null
+    }
+
+    const toolCalls: OpenAI.Chat.Completions.ChatCompletionMessageToolCall[] = []
+    let isFinished = false
+    let lastUsageInfo: any = null
+
+    /**
+     * 统一的完成信号发送逻辑
+     * - 有 finish_reason 时
+     * - 无 finish_reason 但是流正常结束时
+     */
+    const emitCompletionSignals = (controller: TransformStreamDefaultController<GenericChunk>) => {
+      if (isFinished) return
+
+      if (toolCalls.length > 0) {
+        controller.enqueue({
+          type: ChunkType.MCP_TOOL_CREATED,
+          tool_calls: toolCalls
+        })
+      }
+
+      const usage = lastUsageInfo || {
+        prompt_tokens: 0,
+        completion_tokens: 0,
+        total_tokens: 0
+      }
+
+      controller.enqueue({
+        type: ChunkType.LLM_RESPONSE_COMPLETE,
+        response: { usage }
+      })
+
+      // 防止重复发送
+      isFinished = true
+    }
+
+    return (context: ResponseChunkTransformerContext) => ({
+      async transform(chunk: OpenAISdkRawChunk, controller: TransformStreamDefaultController<GenericChunk>) {
+        // 持续更新usage信息
+        if (chunk.usage) {
+          lastUsageInfo = {
+            prompt_tokens: chunk.usage.prompt_tokens || 0,
+            completion_tokens: chunk.usage.completion_tokens || 0,
+            total_tokens: (chunk.usage.prompt_tokens || 0) + (chunk.usage.completion_tokens || 0)
+          }
+        }
+
+        // 处理chunk
+        if ('choices' in chunk && chunk.choices && chunk.choices.length > 0) {
+          const choice = chunk.choices[0]
+
+          if (!choice) return
+
+          // 对于流式响应，使用 delta；对于非流式响应，使用 message。
+          // 然而某些 OpenAI 兼容平台在非流式请求时会错误地返回一个空对象的 delta 字段。
+          // 如果 delta 为空对象，应当忽略它并回退到 message，避免造成内容缺失。
+          let contentSource: OpenAISdkRawContentSource | null = null
+          if ('delta' in choice && choice.delta && Object.keys(choice.delta).length > 0) {
+            contentSource = choice.delta
+          } else if ('message' in choice) {
+            contentSource = choice.message
+          }
+
+          if (!contentSource) return
+
+          const webSearchData = collectWebSearchData(chunk, contentSource, context)
+          if (webSearchData) {
+            controller.enqueue({
+              type: ChunkType.LLM_WEB_SEARCH_COMPLETE,
+              llm_web_search: webSearchData
+            })
+          }
+
+          // 处理推理内容 (e.g. from OpenRouter DeepSeek-R1)
+          // @ts-ignore - reasoning_content is not in standard OpenAI types but some providers use it
+          const reasoningText = contentSource.reasoning_content || contentSource.reasoning
+          if (reasoningText) {
+            controller.enqueue({
+              type: ChunkType.THINKING_DELTA,
+              text: reasoningText
+            })
+          }
+
+          // 处理文本内容
+          if (contentSource.content) {
+            controller.enqueue({
+              type: ChunkType.TEXT_DELTA,
+              text: contentSource.content
+            })
+          }
+
+          // 处理工具调用
+          if (contentSource.tool_calls) {
+            for (const toolCall of contentSource.tool_calls) {
+              if ('index' in toolCall) {
+                const { id, index, function: fun } = toolCall
+                if (fun?.name) {
+                  toolCalls[index] = {
+                    id: id || '',
+                    function: {
+                      name: fun.name,
+                      arguments: fun.arguments || ''
+                    },
+                    type: 'function'
+                  }
+                } else if (fun?.arguments) {
+                  toolCalls[index].function.arguments += fun.arguments
+                }
+              } else {
+                toolCalls.push(toolCall)
+              }
+            }
+          }
+
+          // 处理finish_reason，发送流结束信号
+          if ('finish_reason' in choice && choice.finish_reason) {
+            Logger.debug(`[OpenAIApiClient] Stream finished with reason: ${choice.finish_reason}`)
+            const webSearchData = collectWebSearchData(chunk, contentSource, context)
+            if (webSearchData) {
+              controller.enqueue({
+                type: ChunkType.LLM_WEB_SEARCH_COMPLETE,
+                llm_web_search: webSearchData
+              })
+            }
+            emitCompletionSignals(controller)
+          }
+        }
+      },
+
+      // 流正常结束时，检查是否需要发送完成信号
+      flush(controller) {
+        if (isFinished) return
+
+        Logger.debug('[OpenAIApiClient] Stream ended without finish_reason, emitting fallback completion signals')
+        emitCompletionSignals(controller)
+      }
+    })
+  }
+}