Compare commits

...

95 Commits

Author SHA1 Message Date
suyao
4d77202afd filter: copilot 2025-12-01 13:50:01 +08:00
suyao
8c9d79a7d4 gitignore 2025-12-01 13:45:52 +08:00
suyao
fb9a8e7e2c fix: params map 2025-12-01 02:45:27 +08:00
suyao
14cc38a626 Merge remote-tracking branch 'origin/main' into feat/proxy-api-server 2025-12-01 01:59:49 +08:00
suyao
a231952969 feat: Add model exclusion logic for the Azure OpenAI provider and update the tool call model filter. 2025-12-01 01:59:41 +08:00
suyao
874d69291f refactor: import 2025-12-01 01:33:55 +08:00
suyao
4a913fcef7 Refactor: Remove old cache implementation and integrate new reasoning cache service
- Deleted the old ReasoningCache class and its instance.
- Introduced CacheService for managing reasoning caches.
- Updated unified-messages service to utilize new googleReasoningCache and openRouterReasoningCache.
- Added AiSdkToAnthropicSSE adapter to handle streaming events and integrate with new cache service.
- Reorganized shared adapters to include the new AiSdkToAnthropicSSE adapter.
- Created openrouter adapter with detailed reasoning schemas for better type safety and validation.
2025-12-01 00:36:55 +08:00
kangfenmao
b3a58ec321 chore: update release notes for v1.7.1 2025-11-30 19:57:44 +08:00
Phantom
0097ca80e2 docs: improve CLAUDE.md PR workflow guidelines (#11548)
docs: update CLAUDE.md with PR workflow details

Add critical section about Pull Request workflow requirements including reading the template, following all sections, never skipping, and proper formatting
2025-11-30 18:39:47 +08:00
suyao
4c1466cd27 fix: update 'anthropic-beta' header and add authorization for longcat provider 2025-11-30 18:35:58 +08:00
Phantom
d968df4612 fix(ApiService): properly handle and throw stream errors in API check (#11577)
* fix(ApiService): handle stream errors properly in checkApi

Ensure stream errors are properly caught and thrown when checking API availability

* docs(types): add type safety comment for ResponseError

Add FIXME comment highlighting weak type safety in ResponseError type
2025-11-30 18:34:56 +08:00
Copilot
2bd680361a fix: set CLAUDE_CONFIG_DIR to avoid path encoding issues on Windows with non-ASCII usernames (#11550)
* Initial plan

* fix: set CLAUDE_CONFIG_DIR to avoid path encoding issues on Windows with non-ASCII usernames

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>
2025-11-30 17:58:37 +08:00
SuYao
cc676d4bef fix: improve BashTool command display and enhance ToolTitle layout (#11572)
* fix: improve BashTool command display and enhance ToolTitle layout

* style(ant.css): fix overflow in collapse header text

* fix(i18n): translate toolPendingFallback in multiple languages

---------

Co-authored-by: icarus <eurfelux@gmail.com>
2025-11-30 17:58:23 +08:00
Phantom
3b1155b538 fix: make knowledge base tool always visible regardless of sidebar settings (#11553)
* refactor(knowledgeBaseTool): remove unused sidebar icon visibility check

* refactor(Inputbar): remove unused knowledge icon visibility logic

Simplify knowledge base selection by directly using assistant.knowledge_bases
2025-11-30 17:51:17 +08:00
Phantom
03ff6e1ca6 fix: stabilize home scroll behavior (#11576)
* feat(dom): extend scrollIntoView with Chromium-specific options

Add ChromiumScrollIntoViewOptions interface to support additional scroll container options

* refactor(hooks): optimize timer and scroll position hooks

- Use useMemo for scrollKey in useScrollPosition to avoid unnecessary recalculations
- Refactor useTimer to use useCallback for all functions to prevent unnecessary recreations
- Reorganize function order and improve cleanup logic in useTimer

* fix: stabilize home scroll behavior

* Update src/renderer/src/pages/home/Messages/ChatNavigation.tsx

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* fix(utils/dom): add null check for element in scrollIntoView

Prevent potential runtime errors by gracefully handling falsy elements with a warning log

* fix(hooks): use ref for scroll key to avoid stale closure

* fix(useScrollPosition): add cleanup for scroll handler to prevent memory leaks

---------

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
2025-11-30 17:31:32 +08:00
suyao
350519ac0a Merge remote-tracking branch 'origin/main' into feat/proxy-api-server 2025-11-30 16:47:11 +08:00
Phantom
706fac898a fix(i18n): clarify image-generation endpoint type as OpenAI-based (#11554)
* fix(i18n): remove image-generation translations and clarify endpoint type

Update English locale to specify OpenAI for image generation
Add comments to clarify image-generation endpoint type relationship

* fix(i18n): correct portuguese translations in pt-pt.json
2025-11-30 15:39:09 +08:00
Copilot
f5c144404d fix: persist inputbar text using global variable cache to prevent loss on tab switch (#11558)
* fix: implement draft persistence using CacheService in Inputbar components

* fix: enhance draft persistence in Inputbar components using CacheService

* fix: update cache handling for mentioned models in Inputbar component

* fix: improve validation of cached models in Inputbar component

---------

Co-authored-by: suyao <sy20010504@gmail.com>
2025-11-30 15:37:45 +08:00
Phantom
50a217a638 fix: separate undefined vs none reasoning effort (#11562)
fix(reasoning): handle reasoning effort none case properly

separate undefined and none reasoning effort cases
clean up redundant model checks and add fallback logging
2025-11-30 15:37:18 +08:00
Phantom
444c13e1e3 fix: correct trace token usage (#11575)
fix: correct ai sdk token usage mapping
2025-11-30 15:35:23 +08:00
Phantom
255b19d6ee fix(model): resolve doubao provider model inference issue (#11552)
Fix the issue where doubao provider could not infer models using the name field.
Refactored `isDeepSeekHybridInferenceModel` to use `withModelIdAndNameAsId`
utility function to check both model.id and model.name, avoiding duplicate
calls in `isReasoningModel`.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude <noreply@anthropic.com>
2025-11-30 15:14:12 +08:00
suyao
3989229f61 feat: enhance API version handling and cache functionality
- Updated reasoning cache to use tool-specific keys for better organization.
- Added methods to list cache keys and entries.
- Improved API version regex patterns for more accurate matching.
- Refactored API host formatting to handle leading/trailing whitespace and slashes.
- Added functions to extract and remove trailing API version segments from URLs.
2025-11-30 07:10:10 +08:00
suyao
c6c7c240a3 Merge remote-tracking branch 'origin/main' into feat/proxy-api-server 2025-11-30 02:27:07 +08:00
Phantom
f1f4831157 fix: prevent NaN thinking timers (#11556)
* fix: prevent NaN thinking timers

* test: cover thinking timer fallback and cleanup
2025-11-29 20:29:47 +08:00
suyao
35cfc7c517 feat: add sanitizeToolsForAnthropic function to clean tool definitions for Anthropic API 2025-11-29 19:12:56 +08:00
suyao
e255a992cc fix: type check 2025-11-29 17:39:31 +08:00
xerxesliu
876f59d650 fix: resolve copy image failure for JPEG format pictures (#11529)
- Convert all image formats to PNG before writing to clipboard to ensure compatibility
- Refactor handleCopyImage to unify image source handling (Base64, File, URL)
- Add convertImageToPng utility function using canvas API for robust conversion
- Remove fallback logic that attempted to write unsupported JPEG format
2025-11-29 14:40:49 +08:00
Phantom
c23e88ecd1 fix: handle Gemini API version correctly for Cloudflare Gateway URLs (#11543)
* refactor(api): extract version regex into constant for reuse

Move the version matching regex pattern into a module-level constant to improve code reuse and maintainability. The functionality remains unchanged.

* refactor(api): rename regex constant and use dynamic regex construction

Use a string pattern for version regex to allow dynamic construction and improve maintainability. Rename constant to better reflect its purpose.

* feat(api): add getLastApiVersion utility function

Implement a utility function to extract the last API version segment from URLs. This is useful for handling cases where multiple version segments exist in the path and we need to determine the most specific version being used.

Add comprehensive test cases covering various URL patterns and edge cases.

* feat(api): add utility to remove trailing API version from URLs

Add withoutTrailingApiVersion function to clean up URLs by removing version segments
at the end of paths. This helps standardize API endpoint URLs when version is not needed.

* refactor(api): rename isSupportedAPIVerion to supportApiVersion for clarity

* fix(gemini): handle api version dynamically for non-vertex providers

Use getLastApiVersion utility to determine the latest API version for non-vertex providers instead of hardcoding to v1beta

* feat(api): add function to extract trailing API version from URL

Add getTrailingApiVersion utility function to specifically extract API version segments
that appear at the end of URLs. This complements existing version-related utilities
and helps handle cases where we only care about the final version in the path.

* refactor(gemini): use getTrailingApiVersion instead of getLastApiVersion

The function name was changed to better reflect its purpose of extracting the trailing API version from the URL. The logic was also simplified and made more explicit.

* refactor(api): remove unused getLastApiVersion function

The function was removed as it was no longer needed, simplifying the API version handling to only use trailing version detection. The trailing version regex was extracted to a constant for reuse.
2025-11-29 14:37:26 +08:00
Phantom
284d0f99e1 fix(anthropic): comment out CONTEXT_100M_HEADER to handle via user preferences (#11545)
See #11540 and #11397 for context on moving this to assistant settings
2025-11-29 14:33:10 +08:00
defi-failure
13ac5d564a fix: match tool-call chunk with tool id (#11533) 2025-11-28 20:46:52 +08:00
suyao
e8dccf51fe feat: enhance reasoning cache integration and update provider options in unified messages 2025-11-28 16:37:58 +08:00
kangfenmao
4620b71aee chore: update release notes for v1.7.0 2025-11-28 15:55:12 +08:00
suyao
ed769ac4f7 feat: add CherryAI signed fetch wrapper and enhance tool conversion to Zod schema 2025-11-28 15:48:50 +08:00
kangfenmao
1b926178f1 chore: update @openrouter/ai-sdk-provider to version 1.2.8 in package.json and yarn.lock 2025-11-28 14:44:45 +08:00
Phantom
5167c927be fix: preserve openrouter reasoning with web search (#11505)
* feat(options): implement deep merging for provider options

Add deep merge functionality to preserve nested properties when combining provider options. The new implementation handles object merging recursively while maintaining type safety.

* refactor(tsconfig): reorganize include paths in tsconfig files

Clean up and reorder include paths for better maintainability and consistency between tsconfig.node.json and tsconfig.web.json

* test: add aiCore test configuration and script

Add new test configuration for aiCore package and corresponding test script in package.json to enable running tests specifically for the aiCore module.

* fix: format

* fix(aiCore): resolve test failures and update test infrastructure

- Add vitest setup file with global mocks for @cherrystudio/ai-sdk-provider
- Fix context assertions: use 'model' instead of 'modelId' in plugin tests
- Fix error handling tests: update expected error messages to match actual behavior
- Fix streamText tests: use 'maxOutputTokens' instead of 'maxTokens'
- Fix schemas test: update expected provider list to match actual implementation
- Fix mock-responses: use AI SDK v5 format (inputTokens/outputTokens)
- Update vi.mock to use importOriginal for preserving jsonSchema export

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(aiCore): add alias mock for @cherrystudio/ai-sdk-provider in tests

The vi.mock in setup file doesn't work for source code imports.
Use vitest resolve.alias to mock the external package properly.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(aiCore): disable unused-vars warnings in mock file

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(aiCore): use import.meta.url for ESM compatibility in vitest config

__dirname is not available in ESM modules, use fileURLToPath instead.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(aiCore): use absolute paths in vitest config for workspace compatibility

- Use path.resolve for setupFiles and all alias paths
- Extend aiCore vitest.config.ts from root workspace config
- Change aiCore test environment to 'node' instead of 'jsdom'

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* docs(factory): improve mergeProviderOptions documentation

Add detailed explanation of merge behavior with examples

* test(factory): add tests for mergeProviderOptions behavior

Add test cases to verify mergeProviderOptions correctly handles primitive values, arrays, and nested objects during merging

* refactor(tests): clean up mock responses test fixtures

Remove unused mock streaming chunks and error responses to simplify test fixtures
Update warning details structure in mock complete responses

* docs(test): clarify comment in generateImage test

Update comment to use consistent 'model id' terminology instead of 'modelId'

* test(factory): verify array replacement in mergeProviderOptions

---------

Co-authored-by: suyao <sy20010504@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
2025-11-28 13:56:46 +08:00
suyao
95c18d192a feat: add reasoning cache support to AiSdkToAnthropicSSE and update unified-messages integration 2025-11-28 13:42:33 +08:00
suyao
534d27f37e feat: add additional model IDs for OpenAI Responses endpoint in Copilot 2025-11-28 13:37:43 +08:00
suyao
313be4427b Merge remote-tracking branch 'origin/main' into feat/proxy-api-server 2025-11-28 13:36:33 +08:00
suyao
9d34098a53 feat: enhance provider configuration and error handling for AI SDK integration 2025-11-28 13:36:29 +08:00
suyao
d367040fd4 feat: implement reasoning cache for improved performance and error handling in AI SDK integration 2025-11-28 13:11:13 +08:00
SuYao
b18c64b725 feat: enhance support for AWS Bedrock and Azure OpenAI providers (#11510)
* feat: enhance support for AWS Bedrock and Azure OpenAI providers

* fix: resolve PR review issues for AWS Bedrock support

- Fix header.ts logic bug: change && to || for Vertex/Bedrock provider check
- Fix regex in reasoning.ts to match AWS Bedrock model format (anthropic.claude-*)
- Add test coverage for AWS Bedrock format in isClaude4SeriesModel
- Add Bedrock provider tests including anthropicBeta parameter

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude <noreply@anthropic.com>
2025-11-28 11:00:02 +08:00
Copilot
7ce1590eaf fix: add null checks and type guards to all MessageAgentTools to prevent rendering errors (#11512)
* Initial plan

* fix: add null checks to BashTool to prevent rendering errors

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* fix: add null checks to all MessageAgentTools to prevent rendering errors

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* fix: add Array.isArray checks to prevent map errors on non-array values

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* fix: add typeof checks for string operations to prevent type errors

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* refactor: remove redundant typeof string checks for typed outputs

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>
2025-11-28 10:12:21 +08:00
suyao
356e828422 feat: enhance AI SDK integration with middleware support and improve message handling 2025-11-28 04:12:18 +08:00
suyao
ce25001590 feat: add shared AI SDK middlewares and refactor middleware handling 2025-11-28 01:27:20 +08:00
SuYao
77a9504f74 Fix/condition OpenAI settings (#11509)
* fix(provider): update service tier support logic for OpenAI and Azure providers

* fix(settings): enhance OpenAI settings visibility logic with verbosity support
2025-11-27 22:45:43 +08:00
suyao
77c1b77113 refactor: extract shared token counting logic in messages routes
Extract duplicated token estimation code from both count_tokens endpoints
into a shared `estimateTokenCount` function to improve maintainability.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-27 22:41:18 +08:00
SuYao
bf35902696 fix(mcp): ensure tool uniqueness by using tool IDs for multiple server instances (#11508) 2025-11-27 22:35:24 +08:00
suyao
f163c4d3ee fix: resolve PR review issues for Proxy API Server
- Fix tool result content bug: return `values` array instead of empty array
- Fix empty message bug: skip pushing user/assistant messages when content is empty
- Expand provider support: remove type restrictions to support all AI SDK providers
- Add missing alias for @cherrystudio/ai-sdk-provider in main process config

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-27 22:28:44 +08:00
Phantom
0d12b5fbc2 fix(SelectModelPopup): memoize adapted models to avoid unnecessary updates (#11506)
fix(SelectModelPopup): memoize adapted models to avoid unnecessary update
2025-11-27 22:22:04 +08:00
suyao
0f6ec3e061 feat: add new aliases for ai-core provider and core 2025-11-27 22:02:50 +08:00
Copilot
1746e8b21f Fix MCP server confusion when multiple instances of the same server are configured (#10897)
* Initial plan

* Fix MCP server confusion by making tool IDs unique with serverId

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* Run yarn format to fix code formatting

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* Fix unit test: allow dash separator in tool names

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* Fix edge cases: preserve suffix on truncation, handle non-alphanumeric serverId

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>
2025-11-27 21:50:51 +08:00
suyao
5d1d2b7a9b Merge remote-tracking branch 'origin/main' into feat/proxy-api-server 2025-11-27 21:34:24 +08:00
suyao
15c0a3881c fix: test 2025-11-27 21:28:24 +08:00
suyao
dad9cc95ad fix: typecheck 2025-11-27 21:23:28 +08:00
suyao
f02c0fe962 fix: typecheck 2025-11-27 21:19:01 +08:00
suyao
4c4102da20 feat: update agentModelFilter to exclude generate image models 2025-11-27 21:10:26 +08:00
suyao
2a1adfe322 feat: add ppio 2025-11-27 21:00:52 +08:00
suyao
36ed062b84 fix: test 2025-11-27 20:55:36 +08:00
suyao
f225fbe3e3 feat: Implement direct processing for Anthropic SDK and refactor message handling 2025-11-27 20:43:47 +08:00
xerxesliu
0836eef1a6 fix: store JSON custom parameters as strings instead of objects (#11501) (#11503)
Previously, JSON-type custom parameters were incorrectly parsed and stored
as objects in the UI layer, causing API requests to fail when getCustomParameters()
attempted to JSON.parse() an already-parsed object.

Changes:
- AssistantModelSettings.tsx: Remove JSON.parse() in onChange handler, store as string
- reasoning.ts: Add comments explaining JSON parsing flow
- BaseApiClient.ts: Add comments for legacy API clients
2025-11-27 20:22:27 +08:00
fullex
d0bd10190d feat(test): e2e framework (#11494)
* feat(test): e2e framework

Add Playwright-based e2e testing framework for Electron app with:
- Custom fixtures for electronApp and mainWindow
- Page Object Model (POM) pattern implementation
- 15 example test cases covering app launch, navigation, settings, and chat
- Comprehensive README for humans and AI assistants

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* refactor(tests): update imports and improve code readability

- Changed imports from 'import { Page, Locator }' to 'import type { Locator, Page }' for better type clarity across multiple page files.
- Reformatted waitFor calls in ChatPage and HomePage for improved readability.
- Updated index.ts to correct the export order of ChatPage and SidebarPage.
- Minor adjustments in electron.fixture.ts and electron-app.ts for consistency in import statements.

These changes enhance the maintainability and clarity of the test codebase.

* chore: update linting configuration to include tests directory

- Added 'tests/**' to the ignore patterns in .oxlintrc.json and eslint.config.mjs to ensure test files are not linted.
- Minor adjustment in electron.fixture.ts to improve the fixture definition.

These changes streamline the linting process and enhance code organization.

* fix(test): select main window by title to fix flaky e2e tests on Mac

On Mac, the app may create miniWindow for QuickAssistant alongside mainWindow.
Using firstWindow() could randomly select the wrong window, causing test failures.
Now we wait for the window with title "Cherry Studio" to ensure we get the main window.

Also removed unused electron-app.ts utility file.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude <noreply@anthropic.com>
2025-11-27 19:52:31 +08:00
suyao
ccfb9423e0 chore: format 2025-11-27 19:31:59 +08:00
suyao
192357a32e feat: Enhance thinking block management and tool conversion in unified messages 2025-11-27 19:19:04 +08:00
Phantom
d8191bd4fb refactor: improve verbosity configuration with type-safe validators (#11463)
* refactor(models): improve verbosity level handling for GPT-5 models

Replace hardcoded verbosity configuration with validator functions
Add support for GPT-5.1 series models

* test(models): restructure model utility tests into logical groups

Improve test organization by grouping related test cases under descriptive describe blocks for better maintainability and readability. Each model utility function now has its own dedicated test section with clear subcategories for different behaviors.

* fix: add null check for model in getModelSupportedVerbosity

Handle null model case defensively by returning default verbosity

* refactor(config): remove redundant as const from MODEL_SUPPORTED_VERBOSITY array

* refactor(models): simplify validator function in MODEL_SUPPORTED_VERBOSITY

* test(model utils): add tests for undefined/null input handling

* fix(models): handle undefined/null input in getModelSupportedVerbosity

Remove ts-expect-error comments and update type signature to explicitly handle undefined/null inputs. Also add support for GPT-5.1 series models.

* test(models): add test case for gpt-5-pro variant model
2025-11-27 17:22:33 +08:00
suyao
a5e7aa1342 feat: Implement shared provider utilities and API host formatting
- Added provider API host formatting utilities to handle differences between Cherry Studio and AI SDK.
- Introduced functions for formatting provider API hosts, including support for Azure OpenAI and Vertex AI.
- Created a simple API key rotator for managing API key rotation.
- Developed shared provider initialization and mapping utilities for resolving provider IDs.
- Implemented AI SDK configuration utilities for converting Cherry Studio providers to AI SDK configurations.
- Added support for various providers including OpenRouter, Google Vertex AI, and Amazon Bedrock.
- Enhanced error handling and logging in the unified messages service for better debugging.
- Introduced functions for streaming and generating unified messages using AI SDK.
2025-11-27 15:30:19 +08:00
fullex
d15571c727 fix(code-tools): support Chinese paths and validate directory existence (#11489)
- Add `chcp 65001` to Windows batch file to switch CMD.exe to UTF-8 code page,
  fixing CLI tool launch failure when working directory contains Chinese or
  other non-ASCII characters
- Add directory existence validation before launching terminal to provide
  immediate error feedback instead of delayed failure

Closes #11483

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude <noreply@anthropic.com>
2025-11-27 14:05:14 +08:00
MyPrototypeWhat
a2f67dddb6 fix: resolve readonly property error in assistant preset settings (#11491)
When updating assistant preset settings, if agent.settings was undefined,
it was assigned the DEFAULT_ASSISTANT_SETTINGS object directly. Since this
object is defined with `as const`, it is readonly and subsequent property
assignments would fail with "Cannot assign to read only property".

Fixed by creating a shallow copy of DEFAULT_ASSISTANT_SETTINGS instead of
referencing it directly.

Closes #11490

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude <noreply@anthropic.com>
2025-11-27 13:41:33 +08:00
fullex
8f00321a60 fix: inconsistent text color in release notes last line (#11480)
Move color and font-size styles from p selector to container level
in UpdateNotesWrapper. This ensures all content (including li elements
not wrapped in p tags) uses consistent color.

The issue occurred because .replace(/\n/g, '\n\n') creates a "loose list"
in Markdown where most list items get wrapped in <p> tags, but the last
item (without trailing newline) may not, causing it to inherit a different
color from the parent .markdown class.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude <noreply@anthropic.com>
2025-11-27 10:31:34 +08:00
Apine
eb4670c22c docs: correct the links on the readme (#11477) 2025-11-26 21:17:25 +08:00
fullex
c0beab0f8a chore: update release notes for v1.7.0-rc.3
- Updated version to 1.7.0-rc.3 in package.json
- Added new features including support for Silicon provider and AIHubMix
- Consolidated bug fixes related to providers, models, UI, and settings
- Improved SDK integration with upgraded dependencies
2025-11-26 21:09:27 +08:00
chenxue
97519d96d7 feat(aihubmix): support nano banana (#11476)
support nano banana
2025-11-26 20:51:52 +08:00
Phantom
cbf1d461f0 fix(i18n): clean up translation tags and untranslated strings (#11471)
fix(i18n): update translation strings in ja-jp and ru-ru files

Remove unnecessary translate_input tags and fix incorrect translations
2025-11-26 20:08:04 +08:00
SuYao
bed55c418d fix: silicon provider code list (#11474) 2025-11-26 19:59:57 +08:00
Copilot
82ef4a32eb Fix Poe API reasoning parameters for GPT-5 and reasoning models (#11379)
* Initial plan

* feat: Add proper Poe API reasoning parameters support for GPT-5 and other models

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* test: Add comprehensive tests for Poe API reasoning support

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* fix: Add missing isGPT5SeriesModel import in reasoning.ts

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* fix: Use correct extra_body format for Poe API reasoning parameters

Per Poe API documentation, custom bot parameters like reasoning_effort
and thinking_budget should be passed directly in extra_body, not as
nested structures.

Changed from:
- reasoning_effort: 'low' -> extra_body: { reasoning_effort: 'low' }
- thinking: { type: 'enabled', budget_tokens: X } -> extra_body: { thinking_budget: X }
- extra_body: { google: { thinking_config: {...} } } -> extra_body: { thinking_budget: X }

Updated tests to match the corrected implementation.

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* fix: Update reasoning parameters and improve type definitions for GPT-5 support

* fix lint

* docs

* fix(reasoning): handle edge cases for models without token limit configuration

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>
Co-authored-by: suyao <sy20010504@gmail.com>
2025-11-26 19:56:31 +08:00
槑囿脑袋
79f75843a7 fix: get quota and quota tips (#11472)
Co-authored-by: Claude <noreply@anthropic.com>
2025-11-26 19:53:59 +08:00
Phantom
91f0c47b33 fix(anthropic): prevent duplicate /v1 in API endpoints (#11467)
* fix(anthropic): prevent duplicate /v1 in API endpoints

Anthropic SDK automatically appends /v1 to endpoints, so we should not add it in our formatting. This change ensures URLs are correctly formatted without duplicate path segments.

* fix(anthropic): strip /v1 suffix in getSdkClient to prevent duplicate in models endpoint

The issue was:
- AI SDK (for chat) needs baseURL with /v1 suffix
- Anthropic SDK (for listModels) automatically appends /v1 to all endpoints

Solution:
- Keep /v1 in formatProviderApiHost for AI SDK compatibility
- Strip /v1 in getSdkClient before passing to Anthropic SDK
- This ensures chat works correctly while preventing /v1/v1/models duplication

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(anthropic): correct preview URL to match actual request behavior

The preview now correctly shows:
- Input: https://api.siliconflow.cn/v2
- Preview: https://api.siliconflow.cn/v2/messages (was incorrectly showing /v2/v1/messages)
- Actual: https://api.siliconflow.cn/v2/messages

This matches the actual behavior where getSdkClient strips /v1 suffix before
passing to Anthropic SDK, which then appends /v1/messages.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(anthropic): strip all API version suffixes, not just /v1

The Anthropic SDK always appends /v1 to endpoints, regardless of the baseURL.
Previously we only stripped /v1 suffix, causing issues with custom versions like /v2.

Now we strip all version suffixes (/v1, /v2, /v1beta, etc.) before passing to Anthropic SDK.

Examples:
- Input: https://api.siliconflow.cn/v2/
- After strip: https://api.siliconflow.cn
- Actual request: https://api.siliconflow.cn/v1/messages 

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(anthropic): correct preview to show AI SDK behavior, not Anthropic SDK

The preview was showing the wrong URL because it was reflecting Anthropic SDK behavior
(which strips versions and uses /v1), but checkApi and chat use AI SDK which preserves
the user's version path.

Now preview correctly shows:
- Input: https://api.siliconflow.cn/v2/
- AI SDK (checkApi/chat): https://api.siliconflow.cn/v2/messages 
- Preview: https://api.siliconflow.cn/v2/messages 

Note: Anthropic SDK (for listModels) still strips versions to use /v1/models,
but this is not shown in preview since it's a different code path.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* refactor(checkApi): remove unnecessary legacy fallback

The legacy fallback logic in checkApi was:
1. Complex and hard to maintain
2. Never actually triggered in practice for Modern SDK supported providers
3. Could cause duplicate API requests

Since Modern AI SDK now handles all major providers correctly,
we can simplify by directly throwing errors instead of falling back.

This also removes unused imports: AiProvider and CompletionsParams.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(anthropic): restore version stripping in getSdkClient for Anthropic SDK

The Anthropic SDK (used for listModels) always appends /v1 to endpoints,
so we need to strip version suffixes from baseURL to avoid duplication.

This only affects Anthropic SDK operations (like listModels).
AI SDK operations (chat/checkApi) use provider.apiHost directly via
providerToAiSdkConfig, which preserves the user's version path.

Examples:
- AI SDK (chat): https://api.siliconflow.cn/v1 -> /v1/messages 
- Anthropic SDK (models): https://api.siliconflow.cn/v1 -> strip v1 -> /v1/models 

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(anthropic): ensure AI SDK gets /v1 in baseURL, strip for Anthropic SDK

The correct behavior is:
1. formatProviderApiHost: Add /v1 to apiHost (for AI SDK compatibility)
2. AI SDK (chat/checkApi): Use apiHost with /v1 -> /v1/messages 
3. Anthropic SDK (listModels): Strip /v1 from baseURL -> SDK adds /v1/models 
4. Preview: Show AI SDK behavior (main use case) -> /v1/messages 

Examples:
- Input: https://api.siliconflow.cn
- Formatted: https://api.siliconflow.cn/v1 (added by formatApiHost)
- AI SDK: https://api.siliconflow.cn/v1/messages 
- Anthropic SDK: https://api.siliconflow.cn (stripped) + /v1/models 
- Preview: https://api.siliconflow.cn/v1/messages 

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* refactor(ai): simplify AiProviderNew initialization and improve docs

Update AiProviderNew constructor to automatically format URLs by default
Add comprehensive documentation explaining constructor behavior and usage

* chore: remove unused play.ts file

* fix(anthropic): strip api version from baseURL to avoid endpoint duplication

---------

Co-authored-by: Claude <noreply@anthropic.com>
2025-11-26 19:26:39 +08:00
SuYao
28dff9dfe3 feat: add silicon provider support for Anthropic API compatibility (#11468)
* feat: add silicon provider support for Anthropic API compatibility

* fix: update handling of ANTHROPIC_BASE_URL for silicon provider compatibility

* fix: update anthropicApiHost for silicon provider to use the correct endpoint

* fix: remove silicon from CLAUDE_OFFICIAL_SUPPORTED_PROVIDERS

* chore: add comment to clarify silicon model fallback logic in CLAUDE_OFFICIAL_SUPPORTED_PROVIDERS
2025-11-26 19:19:34 +08:00
fullex
155930ecf4 chore: update @types/react and @types/react-dom to latest versions
- Bumped @types/react from ^19.0.12 to ^19.2.7
- Bumped @types/react-dom from ^19.0.4 to ^19.2.3
- Updated csstype dependency from ^3.0.2 to ^3.2.2 in yarn.lock

These updates ensure compatibility with the latest React types and improve type definitions.
2025-11-26 16:05:40 +08:00
Shuchen Luo
b6b999b635 fix: add claude-opus-4-5 pattern to THINKING_TOKEN_MAP (#11457)
* fix: add claude-opus-4-5 pattern to THINKING_TOKEN_MAP

Adds missing regex pattern for claude-opus-4-5 models (e.g., claude-opus-4-5-20251101)
to the THINKING_TOKEN_MAP configuration. Without this pattern, the model was not
recognized, causing findTokenLimit() to return undefined and leading to an
AI_InvalidArgumentError when using Google Vertex AI Anthropic provider.

The fix adds the pattern 'claude-opus-4-5.*$': { min: 1024, max: 64_000 } to
match the existing claude-4 thinking token configuration.

Fixes AI_InvalidArgumentError: invalid anthropic provider options caused by
budgetTokens receiving NaN instead of a number.

Signed-off-by: Shuchen Luo (personal linux) <nemo0806@gmail.com>

* refactor: make THINKING_TOKEN_MAP constant private

* fix(reasoning): update claude model token limit regex patterns

- Consolidate claude model regex patterns to be more consistent
- Add comprehensive test cases for various claude model variants
- Ensure case insensitivity and proper handling of edge cases

* fix: format

* feat(models): extend claude model regex patterns to support AWS and GCP formats

Update regex patterns in THINKING_TOKEN_MAP to support additional Claude model ID formats used in AWS Bedrock and GCP Vertex AI
Add comprehensive test cases for new model ID formats and reorganize test suite

* fix: format

---------

Signed-off-by: Shuchen Luo (personal linux) <nemo0806@gmail.com>
Co-authored-by: icarus <eurfelux@gmail.com>
2025-11-26 15:47:14 +08:00
SuYao
0d69eeaccf fix: improve Gemini reasoning and message handling (#11439)
* fix: some bug

* fix/test

* fix: lint

* fix: 添加跳过 Gemini3 思考签名的中间件并更新消息转换逻辑

* fix: comment

* fix: js docs

* fix:id bug

* fix: condition

* fix: Update the user's verbosity setting logic to ensure that supported options are prioritized for use.

* fix: Add support for the 'openai-response' provider type.

* fix: lint
2025-11-26 15:46:52 +08:00
Phantom
ff48ce0a58 docs: enhance CLAUDE.md with quality guidelines (#11464)
* docs: add linting and testing step to completion guidelines

* docs: update CLAUDE.md with PR template guideline
2025-11-26 15:45:43 +08:00
SuYao
a2de7d48be fix: update Azure provider handling in AI SDK integration (#11465) 2025-11-26 15:43:32 +08:00
fullex
d4396b4890 docs: update links in Chinese contributing guide
- Corrected the paths in the Chinese version of the contributing guide to point to the appropriate documentation locations.
2025-11-26 13:21:10 +08:00
fullex
283519f1fd Merge branch 'main' of github.com:CherryHQ/cherry-studio 2025-11-26 13:17:09 +08:00
fullex
bb41709ce8 docs: update docs directory structure
- Updated links in CONTRIBUTING.md and README.md to point to the correct Chinese documentation paths.
- Removed outdated files including the English and Chinese versions of the branching strategy, contributing guide, and test plan documents.
- Cleaned up references to non-existent documentation in the project structure to streamline the contributor experience.
2025-11-26 13:17:01 +08:00
Copilot
c1f4b5b9b9 Fix: custom parameters for Gemini models (#11456)
* Initial plan

* fix(aiCore): extract AI SDK standard params from custom params for Gemini

Custom parameters like topK, frequencyPenalty, presencePenalty,
stopSequences, and seed should be passed as top-level streamText()
parameters, not in providerOptions. This fixes the issue where these
parameters were being ignored by the AI SDK's @ai-sdk/google module.

Changes:
- Add extractAiSdkStandardParams function to separate standard params
- Update buildProviderOptions to return both providerOptions and standardParams
- Update buildStreamTextParams to spread standardParams into params object
- Update tests to reflect new return structure

Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>

* refactor(aiCore): remove extractAiSdkStandardParams function and its tests, streamline parameter extraction logic

* chore: type

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: DeJeune <67425183+DeJeune@users.noreply.github.com>
Co-authored-by: suyao <sy20010504@gmail.com>
2025-11-26 13:16:58 +08:00
SuYao
5fb59d21ec fix: header merging logic via chore ai-sdk (#11443)
* fix: update provider-utils and add patch for header merging logic

* fix: enhance header merging logic to deduplicate values

* fix: handle null values in header merging logic

* chore: update ai-sdk dependencies and remove obsolete patches

- Updated @ai-sdk/amazon-bedrock from 3.0.56 to 3.0.61
- Updated @ai-sdk/anthropic from 2.0.45 to 2.0.49
- Updated @ai-sdk/gateway from 2.0.13 to 2.0.15
- Updated @ai-sdk/google from 2.0.40 to 2.0.43
- Updated @ai-sdk/google-vertex from 3.0.72 to 3.0.79
- Updated @ai-sdk/openai from 2.0.71 to 2.0.72
- Updated @ai-sdk/provider-utils from patch version to 3.0.17
- Removed obsolete patches for @ai-sdk/openai and @ai-sdk/provider-utils
- Added reasoning_content field to OpenAIChat response and chunk schemas
- Enhanced OpenAIChatLanguageModel to handle reasoning content in responses

* chore
2025-11-26 12:31:55 +08:00
Phantom
e8de31ca64 fix: Groq verbosity setting (#11452)
* feat(settings): show OpenAI settings for supported service tier providers

Add support for displaying OpenAI settings when provider supports service tiers.
This includes refactoring the condition check and fixing variable naming consistency.

* fix(settings): set openAI verbosity to undefined by default

* fix(store): bump version to 178 and disable verbosity for groq provider

Add migration to remove verbosity from groq provider and implement provider utility to check verbosity support
Update provider types to include verbosity support flag

* feat(provider): add verbosity option support for providers

Add verbosity parameter support in provider API options settings

* fix(aiCore): check provider support for verbosity before applying

Add provider validation and check for verbosity support to prevent errors when unsupported providers are used with verbosity settings

* feat(settings): add Groq settings group component and translations

add new GroqSettingsGroup component for managing Groq provider settings
update translations for Groq settings in both zh-cn and en-us locales
refactor OpenAISettingsGroup to separate Groq-specific logic

* feat(i18n): add groq settings and verbosity support translations

add translations for groq settings title and verbosity parameter support in multiple languages

* refactor(settings): simplify service tier mode fallback logic

Remove conditional service tier mode fallback and use provider-specific defaults directly

* fix(provider): remove redundant system provider check in verbosity support

* test(provider): add tests for verbosity support detection

* fix(OpenAISettingsGroup): add endpoint_type check for showSummarySetting condition

Add model.endpoint_type check to properly determine when to show summary setting for OpenAI models

* refactor(selector): simplify selector option types and add utility functions

remove undefined and null from selector option types
add utility functions to convert between option values and real values
update groq and openai settings groups to use new utilities
add new translation for "ignore" option

* fix(ApiOptionsSettings): correct checked state for verbosity toggle

* feat(i18n): add "ignore" translation for multiple languages

* refactor(groq): remove unused model prop and related checks

Clean up GroqSettingsGroup component by removing unused model prop and unnecessary service tier checks
2025-11-25 23:29:03 +08:00
Phantom
69d31a1e2b fix(models): qwen-mt-flash supports text delta (#11448)
refactor(models): improve text delta support check for qwen-mt models

Replace direct qwen-mt model check with regex pattern matching
Add comprehensive test cases for isNotSupportTextDeltaModel
Update all references to use new function name
2025-11-25 22:22:18 +08:00
fullex
fd3b7f717d fix: correct updateAssistantPreset reducer to properly update preset (#11453)
The previous implementation used `a = preset` inside forEach, which only
reassigns the local variable and doesn't actually update the array element.

Changed to use findIndex + direct array assignment to properly update
the preset in the state.

Fixes #11451

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude <noreply@anthropic.com>
2025-11-25 20:59:56 +08:00
LiuVaayne
bcd7bc9f2d ⬆️ chore: upgrade @anthropic-ai/claude-agent-sdk to 0.1.53 (#11444)
- Upgrade from 0.1.30 to 0.1.53
- Re-apply fork() patch for Electron IPC compatibility
2025-11-25 18:46:11 +08:00
kangfenmao
4dd92c3ce1 fix: handle optional provider in isSupportedReasoningEffortGrokModel function 2025-11-25 17:22:54 +08:00
SuYao
dc8df98929 fix: websearch button condition (#11440)
fix: button
2025-11-25 13:24:37 +08:00
fullex
0004a8cafe fix: respect enableMaxTokens setting when maxTokens is not configured (#11438)
* fix: respect enableMaxTokens setting when maxTokens is not configured

When enableMaxTokens is disabled, getMaxTokens() should return undefined
to let the API use its own default value, instead of forcing 4096 tokens.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(modelParameters): handle max tokens when feature is disabled

Check if max tokens feature is enabled before returning undefined to ensure proper API behavior

---------

Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: icarus <eurfelux@gmail.com>
2025-11-25 11:12:50 +08:00
defi-failure
1992363580 chore: bump version to 1.7.0-rc.2 (#11429) 2025-11-24 14:46:10 +08:00
245 changed files with 10467 additions and 2718 deletions

2
.gitignore vendored
View File

@@ -73,3 +73,5 @@ test-results
YOUR_MEMORY_FILE_PATH
.sessions/
.next/
*.tsbuildinfo

View File

@@ -11,6 +11,7 @@
"dist/**",
"out/**",
"local/**",
"tests/**",
".yarn/**",
".gitignore",
"scripts/cloudflare-worker.js",

View File

@@ -1,8 +1,8 @@
diff --git a/dist/index.js b/dist/index.js
index dc7b74ba55337c491cdf1ab3e39ca68cc4187884..ace8c90591288e42c2957e93c9bf7984f1b22444 100644
index 51ce7e423934fb717cb90245cdfcdb3dae6780e6..0f7f7009e2f41a79a8669d38c8a44867bbff5e1f 100644
--- a/dist/index.js
+++ b/dist/index.js
@@ -472,7 +472,7 @@ function convertToGoogleGenerativeAIMessages(prompt, options) {
@@ -474,7 +474,7 @@ function convertToGoogleGenerativeAIMessages(prompt, options) {
// src/get-model-path.ts
function getModelPath(modelId) {
@@ -12,10 +12,10 @@ index dc7b74ba55337c491cdf1ab3e39ca68cc4187884..ace8c90591288e42c2957e93c9bf7984
// src/google-generative-ai-options.ts
diff --git a/dist/index.mjs b/dist/index.mjs
index 8390439c38cb7eaeb52080862cd6f4c58509e67c..a7647f2e11700dff7e1c8d4ae8f99d3637010733 100644
index f4b77e35c0cbfece85a3ef0d4f4e67aa6dde6271..8d2fecf8155a226006a0bde72b00b6036d4014b6 100644
--- a/dist/index.mjs
+++ b/dist/index.mjs
@@ -478,7 +478,7 @@ function convertToGoogleGenerativeAIMessages(prompt, options) {
@@ -480,7 +480,7 @@ function convertToGoogleGenerativeAIMessages(prompt, options) {
// src/get-model-path.ts
function getModelPath(modelId) {

View File

@@ -1,5 +1,5 @@
diff --git a/dist/index.js b/dist/index.js
index 7481f3b3511078068d87d03855b568b20bb86971..8ac5ec28d2f7ad1b3b0d3f8da945c75674e59637 100644
index bf900591bf2847a3253fe441aad24c06da19c6c1..c1d9bb6fefa2df1383339324073db0a70ea2b5a2 100644
--- a/dist/index.js
+++ b/dist/index.js
@@ -274,6 +274,7 @@ var openaiChatResponseSchema = (0, import_provider_utils3.lazyValidator)(

View File

@@ -1,8 +1,8 @@
diff --git a/sdk.mjs b/sdk.mjs
index 8cc6aaf0b25bcdf3c579ec95cde12d419fcb2a71..3b3b8beaea5ad2bbac26a15f792058306d0b059f 100755
index bf429a344b7d59f70aead16b639f949b07688a81..f77d50cc5d3fb04292cb3ac7fa7085d02dcc628f 100755
--- a/sdk.mjs
+++ b/sdk.mjs
@@ -6213,7 +6213,7 @@ function createAbortController(maxListeners = DEFAULT_MAX_LISTENERS) {
@@ -6250,7 +6250,7 @@ function createAbortController(maxListeners = DEFAULT_MAX_LISTENERS) {
}
// ../src/transport/ProcessTransport.ts
@@ -11,16 +11,20 @@ index 8cc6aaf0b25bcdf3c579ec95cde12d419fcb2a71..3b3b8beaea5ad2bbac26a15f79205830
import { createInterface } from "readline";
// ../src/utils/fsOperations.ts
@@ -6505,14 +6505,11 @@ class ProcessTransport {
@@ -6619,18 +6619,11 @@ class ProcessTransport {
const errorMessage = isNativeBinary(pathToClaudeCodeExecutable) ? `Claude Code native binary not found at ${pathToClaudeCodeExecutable}. Please ensure Claude Code is installed via native installer or specify a valid path with options.pathToClaudeCodeExecutable.` : `Claude Code executable not found at ${pathToClaudeCodeExecutable}. Is options.pathToClaudeCodeExecutable set?`;
throw new ReferenceError(errorMessage);
}
- const isNative = isNativeBinary(pathToClaudeCodeExecutable);
- const spawnCommand = isNative ? pathToClaudeCodeExecutable : executable;
- const spawnArgs = isNative ? [...executableArgs, ...args] : [...executableArgs, pathToClaudeCodeExecutable, ...args];
- this.logForDebugging(isNative ? `Spawning Claude Code native binary: ${spawnCommand} ${spawnArgs.join(" ")}` : `Spawning Claude Code process: ${spawnCommand} ${spawnArgs.join(" ")}`);
+ this.logForDebugging(`Forking Claude Code Node.js process: ${pathToClaudeCodeExecutable} ${args.join(" ")}`);
const stderrMode = env.DEBUG || stderr ? "pipe" : "ignore";
- const spawnMessage = isNative ? `Spawning Claude Code native binary: ${spawnCommand} ${spawnArgs.join(" ")}` : `Spawning Claude Code process: ${spawnCommand} ${spawnArgs.join(" ")}`;
- logForSdkDebugging(spawnMessage);
- if (stderr) {
- stderr(spawnMessage);
- }
+ logForSdkDebugging(`Forking Claude Code Node.js process: ${pathToClaudeCodeExecutable} ${args.join(" ")}`);
const stderrMode = env.DEBUG_CLAUDE_AGENT_SDK || stderr ? "pipe" : "ignore";
- this.child = spawn(spawnCommand, spawnArgs, {
+ this.child = fork(pathToClaudeCodeExecutable, args, {
cwd,

View File

@@ -10,8 +10,18 @@ This file provides guidance to AI coding assistants when working with code in th
- **Log centrally**: Route all logging through `loggerService` with the right context—no `console.log`.
- **Research via subagent**: Lean on `subagent` for external docs, APIs, news, and references.
- **Always propose before executing**: Before making any changes, clearly explain your planned approach and wait for explicit user approval to ensure alignment and prevent unwanted modifications.
- **Lint, test, and format before completion**: Coding tasks are only complete after running `yarn lint`, `yarn test`, and `yarn format` successfully.
- **Write conventional commits**: Commit small, focused changes using Conventional Commit messages (e.g., `feat:`, `fix:`, `refactor:`, `docs:`).
## Pull Request Workflow (CRITICAL)
When creating a Pull Request, you MUST:
1. **Read the PR template first**: Always read `.github/pull_request_template.md` before creating the PR
2. **Follow ALL template sections**: Structure the `--body` parameter to include every section from the template
3. **Never skip sections**: Include all sections even if marking them as N/A or "None"
4. **Use proper formatting**: Match the template's markdown structure exactly (headings, checkboxes, code blocks)
## Development Commands
- **Install**: `yarn install` - Install all project dependencies

View File

@@ -1,4 +1,4 @@
[中文](docs/CONTRIBUTING.zh.md) | [English](CONTRIBUTING.md)
[中文](docs/zh/guides/contributing.md) | [English](CONTRIBUTING.md)
# Cherry Studio Contributor Guide
@@ -32,7 +32,7 @@ To help you get familiar with the codebase, we recommend tackling issues tagged
### Testing
Features without tests are considered non-existent. To ensure code is truly effective, relevant processes should be covered by unit tests and functional tests. Therefore, when considering contributions, please also consider testability. All tests can be run locally without dependency on CI. Please refer to the "Testing" section in the [Developer Guide](docs/dev.md).
Features without tests are considered non-existent. To ensure code is truly effective, relevant processes should be covered by unit tests and functional tests. Therefore, when considering contributions, please also consider testability. All tests can be run locally without dependency on CI. Please refer to the "Testing" section in the [Developer Guide](docs/zh/guides/development.md).
### Automated Testing for Pull Requests
@@ -60,7 +60,7 @@ Maintainers are here to help you implement your use case within a reasonable tim
### Participating in the Test Plan
The Test Plan aims to provide users with a more stable application experience and faster iteration speed. For details, please refer to the [Test Plan](docs/testplan-en.md).
The Test Plan aims to provide users with a more stable application experience and faster iteration speed. For details, please refer to the [Test Plan](docs/en/guides/test-plan.md).
### Other Suggestions

View File

@@ -34,7 +34,7 @@
</a>
</h1>
<p align="center">English | <a href="./docs/README.zh.md">中文</a> | <a href="https://cherry-ai.com">Official Site</a> | <a href="https://docs.cherry-ai.com/cherry-studio-wen-dang/en-us">Documents</a> | <a href="./docs/dev.md">Development</a> | <a href="https://github.com/CherryHQ/cherry-studio/issues">Feedback</a><br></p>
<p align="center">English | <a href="./docs/zh/README.md">中文</a> | <a href="https://cherry-ai.com">Official Site</a> | <a href="https://docs.cherry-ai.com/cherry-studio-wen-dang/en-us">Documents</a> | <a href="./docs/en/guides/development.md">Development</a> | <a href="https://github.com/CherryHQ/cherry-studio/issues">Feedback</a><br></p>
<div align="center">
@@ -67,7 +67,7 @@ Cherry Studio is a desktop client that supports multiple LLM providers, availabl
👏 Join [Telegram Group](https://t.me/CherryStudioAI)[Discord](https://discord.gg/wez8HtpxqQ) | [QQ Group(575014769)](https://qm.qq.com/q/lo0D4qVZKi)
❤️ Like Cherry Studio? Give it a star 🌟 or [Sponsor](docs/sponsor.md) to support the development!
❤️ Like Cherry Studio? Give it a star 🌟 or [Sponsor](docs/zh/guides/sponsor.md) to support the development!
# 🌠 Screenshot
@@ -175,7 +175,7 @@ We welcome contributions to Cherry Studio! Here are some ways you can contribute
6. **Community Engagement**: Join discussions and help users.
7. **Promote Usage**: Spread the word about Cherry Studio.
Refer to the [Branching Strategy](docs/branching-strategy-en.md) for contribution guidelines
Refer to the [Branching Strategy](docs/en/guides/branching-strategy.md) for contribution guidelines
## Getting Started

81
docs/README.md Normal file
View File

@@ -0,0 +1,81 @@
# Cherry Studio Documentation / 文档
This directory contains the project documentation in multiple languages.
本目录包含多语言项目文档。
---
## Languages / 语言
- **[中文文档](./zh/README.md)** - Chinese Documentation
- **English Documentation** - See sections below
---
## English Documentation
### Guides
| Document | Description |
|----------|-------------|
| [Development Setup](./en/guides/development.md) | Development environment setup |
| [Branching Strategy](./en/guides/branching-strategy.md) | Git branching workflow |
| [i18n Guide](./en/guides/i18n.md) | Internationalization guide |
| [Logging Guide](./en/guides/logging.md) | How to use the logger service |
| [Test Plan](./en/guides/test-plan.md) | Test plan and release channels |
### References
| Document | Description |
|----------|-------------|
| [App Upgrade Config](./en/references/app-upgrade.md) | Application upgrade configuration |
| [CodeBlockView Component](./en/references/components/code-block-view.md) | Code block view component |
| [Image Preview Components](./en/references/components/image-preview.md) | Image preview components |
---
## 中文文档
### 指南 (Guides)
| 文档 | 说明 |
|------|------|
| [开发环境设置](./zh/guides/development.md) | 开发环境配置 |
| [贡献指南](./zh/guides/contributing.md) | 如何贡献代码 |
| [分支策略](./zh/guides/branching-strategy.md) | Git 分支工作流 |
| [测试计划](./zh/guides/test-plan.md) | 测试计划和发布通道 |
| [国际化指南](./zh/guides/i18n.md) | 国际化开发指南 |
| [日志使用指南](./zh/guides/logging.md) | 如何使用日志服务 |
| [中间件开发](./zh/guides/middleware.md) | 如何编写中间件 |
| [记忆功能](./zh/guides/memory.md) | 记忆功能使用指南 |
| [赞助信息](./zh/guides/sponsor.md) | 赞助相关信息 |
### 参考 (References)
| 文档 | 说明 |
|------|------|
| [消息系统](./zh/references/message-system.md) | 消息系统架构和 API |
| [数据库结构](./zh/references/database.md) | 数据库表结构 |
| [服务](./zh/references/services.md) | 服务层文档 (KnowledgeService) |
| [代码执行](./zh/references/code-execution.md) | 代码执行功能 |
| [应用升级配置](./zh/references/app-upgrade.md) | 应用升级配置 |
| [CodeBlockView 组件](./zh/references/components/code-block-view.md) | 代码块视图组件 |
| [图像预览组件](./zh/references/components/image-preview.md) | 图像预览组件 |
---
## Missing Translations / 缺少翻译
The following documents are only available in Chinese and need English translations:
以下文档仅有中文版本,需要英文翻译:
- `guides/contributing.md`
- `guides/memory.md`
- `guides/middleware.md`
- `guides/sponsor.md`
- `references/message-system.md`
- `references/database.md`
- `references/services.md`
- `references/code-execution.md`

View File

Before

Width:  |  Height:  |  Size: 150 KiB

After

Width:  |  Height:  |  Size: 150 KiB

View File

Before

Width:  |  Height:  |  Size: 40 KiB

After

Width:  |  Height:  |  Size: 40 KiB

View File

Before

Width:  |  Height:  |  Size: 35 KiB

After

Width:  |  Height:  |  Size: 35 KiB

View File

Before

Width:  |  Height:  |  Size: 563 KiB

After

Width:  |  Height:  |  Size: 563 KiB

View File

@@ -16,7 +16,7 @@ Cherry Studio implements a structured branching strategy to maintain code qualit
- Only accepts documentation updates and bug fixes
- Thoroughly tested before production deployment
For details about the `testplan` branch used in the Test Plan, please refer to the [Test Plan](testplan-en.md).
For details about the `testplan` branch used in the Test Plan, please refer to the [Test Plan](./test-plan.md).
## Contributing Branches

View File

@@ -18,11 +18,11 @@ The plugin has already been configured in the project — simply install it to g
### Demo
![demo-1](./.assets.how-to-i18n/demo-1.png)
![demo-1](../../assets/images/i18n/demo-1.png)
![demo-2](./.assets.how-to-i18n/demo-2.png)
![demo-2](../../assets/images/i18n/demo-2.png)
![demo-3](./.assets.how-to-i18n/demo-3.png)
![demo-3](../../assets/images/i18n/demo-3.png)
## i18n Conventions

View File

@@ -19,7 +19,7 @@ Users are welcome to submit issues or provide feedback through other channels fo
### Participating in the Test Plan
Developers should submit `PRs` according to the [Contributor Guide](../CONTRIBUTING.md) (and ensure the target branch is `main`). The repository maintainers will evaluate whether the `PR` should be included in the Test Plan based on factors such as the impact of the feature on the application, its importance, and whether broader testing is needed.
Developers should submit `PRs` according to the [Contributor Guide](../../CONTRIBUTING.md) (and ensure the target branch is `main`). The repository maintainers will evaluate whether the `PR` should be included in the Test Plan based on factors such as the impact of the feature on the application, its importance, and whether broader testing is needed.
If the `PR` is added to the Test Plan, the repository maintainers will:

View File

@@ -85,7 +85,7 @@ Main responsibilities:
- **SvgPreview**: SVG image preview
- **GraphvizPreview**: Graphviz diagram preview
All special view components share a common architecture for consistent user experience and functionality. For detailed information about these components and their implementation, see [Image Preview Components Documentation](./ImagePreview-en.md).
All special view components share a common architecture for consistent user experience and functionality. For detailed information about these components and their implementation, see [Image Preview Components Documentation](./image-preview.md).
#### StatusBar

View File

@@ -192,4 +192,4 @@ Image Preview Components integrate seamlessly with CodeBlockView:
- Shared state management
- Responsive layout adaptation
For more information about the overall CodeBlockView architecture, see [CodeBlockView Documentation](./CodeBlockView-en.md).
For more information about the overall CodeBlockView architecture, see [CodeBlockView Documentation](./code-block-view.md).

View File

@@ -1,3 +0,0 @@
# 消息的生命周期
![image](./message-lifecycle.png)

View File

@@ -1,11 +0,0 @@
# 数据库设置字段
此文档包含部分字段的数据类型说明。
## 字段
| 字段名 | 类型 | 说明 |
| ------------------------------ | ------------------------------ | ------------ |
| `translate:target:language` | `LanguageCode` | 翻译目标语言 |
| `translate:source:language` | `LanguageCode` | 翻译源语言 |
| `translate:bidirectional:pair` | `[LanguageCode, LanguageCode]` | 双向翻译对 |

View File

@@ -1,127 +0,0 @@
# messageBlock.ts 使用指南
该文件定义了用于管理应用程序中所有 `MessageBlock` 实体的 Redux Slice。它使用 Redux Toolkit 的 `createSlice``createEntityAdapter` 来高效地处理规范化的状态,并提供了一系列 actions 和 selectors 用于与消息块数据交互。
## 核心目标
- **状态管理**: 集中管理所有 `MessageBlock` 的状态。`MessageBlock` 代表消息中的不同内容单元(如文本、代码、图片、引用等)。
- **规范化**: 使用 `createEntityAdapter``MessageBlock` 数据存储在规范化的结构中(`{ ids: [], entities: {} }`),这有助于提高性能和简化更新逻辑。
- **可预测性**: 提供明确的 actions 来修改状态,并通过 selectors 安全地访问状态。
## 关键概念
- **Slice (`createSlice`)**: Redux Toolkit 的核心 API用于创建包含 reducer 逻辑、action creators 和初始状态的 Redux 模块。
- **Entity Adapter (`createEntityAdapter`)**: Redux Toolkit 提供的工具,用于简化对规范化数据的 CRUD创建、读取、更新、删除操作。它会自动生成 reducer 函数和 selectors。
- **Selectors**: 用于从 Redux store 中派生和计算数据的函数。Selectors 可以被记忆化memoized以提高性能。
## State 结构
`messageBlocks` slice 的状态结构由 `createEntityAdapter` 定义,大致如下:
```typescript
{
ids: string[]; // 存储所有 MessageBlock ID 的有序列表
entities: { [id: string]: MessageBlock }; // 按 ID 存储 MessageBlock 对象的字典
loadingState: 'idle' | 'loading' | 'succeeded' | 'failed'; // (可选) 其他状态,如加载状态
error: string | null; // (可选) 错误信息
}
```
## Actions
该 slice 导出以下 actions (由 `createSlice``createEntityAdapter` 自动生成或自定义)
- **`upsertOneBlock(payload: MessageBlock)`**:
- 添加一个新的 `MessageBlock` 或更新一个已存在的 `MessageBlock`。如果 payload 中的 `id` 已存在,则执行更新;否则执行插入。
- **`upsertManyBlocks(payload: MessageBlock[])`**:
- 添加或更新多个 `MessageBlock`。常用于批量加载数据(例如,加载一个 Topic 的所有消息块)。
- **`removeOneBlock(payload: string)`**:
- 根据提供的 `id` (payload) 移除单个 `MessageBlock`
- **`removeManyBlocks(payload: string[])`**:
- 根据提供的 `id` 数组 (payload) 移除多个 `MessageBlock`。常用于删除消息或清空 Topic 时清理相关的块。
- **`removeAllBlocks()`**:
- 移除 state 中的所有 `MessageBlock` 实体。
- **`updateOneBlock(payload: { id: string; changes: Partial<MessageBlock> })`**:
- 更新一个已存在的 `MessageBlock``payload` 需要包含块的 `id` 和一个包含要更改的字段的 `changes` 对象。
- **`setMessageBlocksLoading(payload: 'idle' | 'loading')`**:
- (自定义) 设置 `loadingState` 属性。
- **`setMessageBlocksError(payload: string)`**:
- (自定义) 设置 `loadingState``'failed'` 并记录错误信息。
**使用示例 (在 Thunk 或其他 Dispatch 的地方):**
```typescript
import { upsertOneBlock, removeManyBlocks, updateOneBlock } from './messageBlock'
import store from './store' // 假设这是你的 Redux store 实例
// 添加或更新一个块
const newBlock: MessageBlock = {
/* ... block data ... */
}
store.dispatch(upsertOneBlock(newBlock))
// 更新一个块的内容
store.dispatch(updateOneBlock({ id: blockId, changes: { content: 'New content' } }))
// 删除多个块
const blockIdsToRemove = ['id1', 'id2']
store.dispatch(removeManyBlocks(blockIdsToRemove))
```
## Selectors
该 slice 导出由 `createEntityAdapter` 生成的基础 selectors并通过 `messageBlocksSelectors` 对象访问:
- **`messageBlocksSelectors.selectIds(state: RootState): string[]`**: 返回包含所有块 ID 的数组。
- **`messageBlocksSelectors.selectEntities(state: RootState): { [id: string]: MessageBlock }`**: 返回块 ID 到块对象的映射字典。
- **`messageBlocksSelectors.selectAll(state: RootState): MessageBlock[]`**: 返回包含所有块对象的数组。
- **`messageBlocksSelectors.selectTotal(state: RootState): number`**: 返回块的总数。
- **`messageBlocksSelectors.selectById(state: RootState, id: string): MessageBlock | undefined`**: 根据 ID 返回单个块对象,如果找不到则返回 `undefined`
**此外,还提供了一个自定义的、记忆化的 selector**
- **`selectFormattedCitationsByBlockId(state: RootState, blockId: string | undefined): Citation[]`**:
- 接收一个 `blockId`
- 如果该 ID 对应的块是 `CITATION` 类型,则提取并格式化其包含的引用信息(来自网页搜索、知识库等),进行去重和重新编号,最后返回一个 `Citation[]` 数组,用于在 UI 中显示。
- 如果块不存在或类型不匹配,返回空数组 `[]`
- 这个 selector 封装了处理不同引用来源Gemini, OpenAI, OpenRouter, Zhipu 等)的复杂逻辑。
**使用示例 (在 React 组件或 `useSelector` 中):**
```typescript
import { useSelector } from 'react-redux'
import { messageBlocksSelectors, selectFormattedCitationsByBlockId } from './messageBlock'
import type { RootState } from './store'
// 获取所有块
const allBlocks = useSelector(messageBlocksSelectors.selectAll)
// 获取特定 ID 的块
const specificBlock = useSelector((state: RootState) => messageBlocksSelectors.selectById(state, someBlockId))
// 获取特定引用块格式化后的引用列表
const formattedCitations = useSelector((state: RootState) => selectFormattedCitationsByBlockId(state, citationBlockId))
// 在组件中使用引用数据
// {formattedCitations.map(citation => ...)}
```
## 集成
`messageBlock.ts` slice 通常与 `messageThunk.ts` 中的 Thunks 紧密协作。Thunks 负责处理异步逻辑(如 API 调用、数据库操作),并在需要时 dispatch `messageBlock` slice 的 actions 来更新状态。例如,当 `messageThunk` 接收到流式响应时,它会 dispatch `upsertOneBlock``updateOneBlock` 来实时更新对应的 `MessageBlock`。同样,删除消息的 Thunk 会 dispatch `removeManyBlocks`
理解 `messageBlock.ts` 的职责是管理**状态本身**,而 `messageThunk.ts` 负责**触发状态变更**的异步流程,这对于维护清晰的应用架构至关重要。

View File

@@ -1,105 +0,0 @@
# messageThunk.ts 使用指南
该文件包含用于管理应用程序中消息流、处理助手交互以及同步 Redux 状态与 IndexedDB 数据库的核心 Thunk Action Creators。主要围绕 `Message``MessageBlock` 对象进行操作。
## 核心功能
1. **发送/接收消息**: 处理用户消息的发送,触发助手响应,并流式处理返回的数据,将其解析为不同的 `MessageBlock`
2. **状态管理**: 确保 Redux store 中的消息和消息块状态与 IndexedDB 中的持久化数据保持一致。
3. **消息操作**: 提供删除、重发、重新生成、编辑后重发、追加响应、克隆等消息生命周期管理功能。
4. **Block 处理**: 动态创建、更新和保存各种类型的 `MessageBlock`(文本、思考过程、工具调用、引用、图片、错误、翻译等)。
## 主要 Thunks
以下是一些关键的 Thunk 函数及其用途:
1. **`sendMessage(userMessage, userMessageBlocks, assistant, topicId)`**
- **用途**: 发送一条新的用户消息。
- **流程**:
- 保存用户消息 (`userMessage`) 及其块 (`userMessageBlocks`) 到 Redux 和 DB。
- 检查 `@mentions` 以确定是单模型响应还是多模型响应。
- 创建助手消息(们)的存根 (Stub)。
- 将存根添加到 Redux 和 DB。
- 将核心处理逻辑 `fetchAndProcessAssistantResponseImpl` 添加到该 `topicId` 的队列中以获取实际响应。
- **Block 相关**: 主要处理用户消息的初始 `MessageBlock` 保存。
2. **`fetchAndProcessAssistantResponseImpl(dispatch, getState, topicId, assistant, assistantMessage)`**
- **用途**: (内部函数) 获取并处理单个助手响应的核心逻辑,被 `sendMessage`, `resend...`, `regenerate...`, `append...` 等调用。
- **流程**:
- 设置 Topic 加载状态。
- 准备上下文消息。
- 调用 `fetchChatCompletion` API 服务。
- 使用 `createStreamProcessor` 处理流式响应。
- 通过各种回调 (`onTextChunk`, `onThinkingChunk`, `onToolCallComplete`, `onImageGenerated`, `onError`, `onComplete` 等) 处理不同类型的事件。
- **Block 相关**:
- 根据流事件创建初始 `UNKNOWN` 块。
- 实时创建和更新 `MAIN_TEXT``THINKING` 块,使用 `throttledBlockUpdate``throttledBlockDbUpdate` 进行节流更新。
- 创建 `TOOL`, `CITATION`, `IMAGE`, `ERROR` 等类型的块。
- 在事件完成时(如 `onTextComplete`, `onToolCallComplete`)将块状态标记为 `SUCCESS``ERROR`,并使用 `saveUpdatedBlockToDB` 保存最终状态。
- 使用 `handleBlockTransition` 管理非流式块(如 `TOOL`, `CITATION`)的添加和状态更新。
3. **`loadTopicMessagesThunk(topicId, forceReload)`**
- **用途**: 从数据库加载指定主题的所有消息及其关联的 `MessageBlock`
- **流程**:
- 从 DB 获取 `Topic` 及其 `messages` 列表。
- 根据消息 ID 列表从 DB 获取所有相关的 `MessageBlock`
- 使用 `upsertManyBlocks` 将块更新到 Redux。
- 将消息更新到 Redux。
- **Block 相关**: 负责将持久化的 `MessageBlock` 加载到 Redux 状态。
4. **删除 Thunks**
- `deleteSingleMessageThunk(topicId, messageId)`: 删除单个消息及其所有 `MessageBlock`
- `deleteMessageGroupThunk(topicId, askId)`: 删除一个用户消息及其所有相关的助手响应消息和它们的所有 `MessageBlock`
- `clearTopicMessagesThunk(topicId)`: 清空主题下的所有消息及其所有 `MessageBlock`
- **Block 相关**: 从 Redux 和 DB 中移除指定的 `MessageBlock`
5. **重发/重新生成 Thunks**
- `resendMessageThunk(topicId, userMessageToResend, assistant)`: 重发用户消息。会重置(清空 Block 并标记为 PENDING所有与该用户消息关联的助手响应然后重新请求生成。
- `resendUserMessageWithEditThunk(topicId, originalMessage, mainTextBlockId, editedContent, assistant)`: 用户编辑消息内容后重发。先更新用户消息的 `MAIN_TEXT` 块内容,然后调用 `resendMessageThunk`
- `regenerateAssistantResponseThunk(topicId, assistantMessageToRegenerate, assistant)`: 重新生成单个助手响应。重置该助手消息(清空 Block 并标记为 PENDING然后重新请求生成。
- **Block 相关**: 删除旧的 `MessageBlock`,并在重新生成过程中创建新的 `MessageBlock`
6. **`appendAssistantResponseThunk(topicId, existingAssistantMessageId, newModel, assistant)`**
- **用途**: 在已有的对话上下文中,针对同一个用户问题,使用新选择的模型追加一个新的助手响应。
- **流程**:
- 找到现有助手消息以获取原始 `askId`
- 创建使用 `newModel` 的新助手消息存根(使用相同的 `askId`)。
- 添加新存根到 Redux 和 DB。
-`fetchAndProcessAssistantResponseImpl` 添加到队列以生成新响应。
- **Block 相关**: 为新的助手响应创建全新的 `MessageBlock`
7. **`cloneMessagesToNewTopicThunk(sourceTopicId, branchPointIndex, newTopic)`**
- **用途**: 将源主题的部分消息(及其 Block克隆到一个**已存在**的新主题中。
- **流程**:
- 复制指定索引前的消息。
- 为所有克隆的消息和 Block 生成新的 UUID。
- 正确映射克隆消息之间的 `askId` 关系。
- 复制 `MessageBlock` 内容,更新其 `messageId` 指向新的消息 ID。
- 更新文件引用计数(如果 Block 是文件或图片)。
- 将克隆的消息和 Block 保存到新主题的 Redux 状态和 DB 中。
- **Block 相关**: 创建 `MessageBlock` 的副本,并更新其 ID 和 `messageId`
8. **`initiateTranslationThunk(messageId, topicId, targetLanguage, sourceBlockId?, sourceLanguage?)`**
- **用途**: 为指定消息启动翻译流程,创建一个初始的 `TRANSLATION` 类型的 `MessageBlock`
- **流程**:
- 创建一个状态为 `STREAMING``TranslationMessageBlock`
- 将其添加到 Redux 和 DB。
- 更新原消息的 `blocks` 列表以包含新的翻译块 ID。
- **Block 相关**: 创建并保存一个占位的 `TranslationMessageBlock`。实际翻译内容的获取和填充需要后续步骤。
## 内部机制和注意事项
- **数据库交互**: 通过 `saveMessageAndBlocksToDB`, `updateExistingMessageAndBlocksInDB`, `saveUpdatesToDB`, `saveUpdatedBlockToDB`, `throttledBlockDbUpdate` 等辅助函数与 IndexedDB (`db`) 交互,确保数据持久化。
- **状态同步**: Thunks 负责协调 Redux Store 和 IndexedDB 之间的数据一致性。
- **队列 (`getTopicQueue`)**: 使用 `AsyncQueue` 确保对同一主题的操作(尤其是 API 请求)按顺序执行,避免竞态条件。
- **节流 (`throttle`)**: 对流式响应中频繁的 Block 更新(文本、思考)使用 `lodash.throttle` 优化性能,减少 Redux dispatch 和 DB 写入次数。
- **错误处理**: `fetchAndProcessAssistantResponseImpl` 内的回调函数(特别是 `onError`)处理流处理和 API 调用中可能出现的错误,并创建 `ERROR` 类型的 `MessageBlock`
开发者在使用这些 Thunks 时,通常需要提供 `dispatch`, `getState` (由 Redux Thunk 中间件注入),以及如 `topicId`, `assistant` 配置对象, 相关的 `Message``MessageBlock` 对象/ID 等参数。理解每个 Thunk 的职责和它如何影响消息及块的状态至关重要。

View File

@@ -1,156 +0,0 @@
# useMessageOperations.ts 使用指南
该文件定义了一个名为 `useMessageOperations` 的自定义 React Hook。这个 Hook 的主要目的是为 React 组件提供一个便捷的接口用于执行与特定主题Topic相关的各种消息操作。它封装了调用 Redux Thunks (`messageThunk.ts`) 和 Actions (`newMessage.ts`, `messageBlock.ts`) 的逻辑,简化了组件与消息数据交互的代码。
## 核心目标
- **封装**: 将复杂的消息操作逻辑(如删除、重发、重新生成、编辑、翻译等)封装在易于使用的函数中。
- **简化**: 让组件可以直接调用这些操作函数,而无需直接与 Redux `dispatch` 或 Thunks 交互。
- **上下文关联**: 所有操作都与传入的 `topic` 对象相关联,确保操作作用于正确的主题。
## 如何使用
在你的 React 函数组件中,导入并调用 `useMessageOperations` Hook并传入当前活动的 `Topic` 对象。
```typescript
import React from 'react';
import { useMessageOperations } from '@renderer/hooks/useMessageOperations';
import type { Topic, Message, Assistant, Model } from '@renderer/types';
interface MyComponentProps {
currentTopic: Topic;
currentAssistant: Assistant;
}
function MyComponent({ currentTopic, currentAssistant }: MyComponentProps) {
const {
deleteMessage,
resendMessage,
regenerateAssistantMessage,
appendAssistantResponse,
getTranslationUpdater,
createTopicBranch,
// ... 其他操作函数
} = useMessageOperations(currentTopic);
const handleDelete = (messageId: string) => {
deleteMessage(messageId);
};
const handleResend = (message: Message) => {
resendMessage(message, currentAssistant);
};
const handleAppend = (existingMsg: Message, newModel: Model) => {
appendAssistantResponse(existingMsg, newModel, currentAssistant);
}
// ... 在组件中使用其他操作函数
return (
<div>
{/* Component UI */}
<button onClick={() => handleDelete('some-message-id')}>Delete Message</button>
{/* ... */}
</div>
);
}
```
## 返回值
`useMessageOperations(topic)` Hook 返回一个包含以下函数和值的对象:
- **`deleteMessage(id: string)`**:
- 删除指定 `id` 的单个消息。
- 内部调用 `deleteSingleMessageThunk`
- **`deleteGroupMessages(askId: string)`**:
- 删除与指定 `askId` 相关联的一组消息(通常是用户提问及其所有助手回答)。
- 内部调用 `deleteMessageGroupThunk`
- **`editMessage(messageId: string, updates: Partial<Message>)`**:
- 更新指定 `messageId` 的消息的部分属性。
- **注意**: 目前主要用于更新 Redux 状态
- 内部调用 `newMessagesActions.updateMessage`
- **`resendMessage(message: Message, assistant: Assistant)`**:
- 重新发送指定的用户消息 (`message`),这将触发其所有关联助手响应的重新生成。
- 内部调用 `resendMessageThunk`
- **`resendUserMessageWithEdit(message: Message, editedContent: string, assistant: Assistant)`**:
- 在用户消息的主要文本块被编辑后,重新发送该消息。
- 会先查找消息的 `MAIN_TEXT` 块 ID然后调用 `resendUserMessageWithEditThunk`
- **`clearTopicMessages(_topicId?: string)`**:
- 清除当前主题(或可选的指定 `_topicId`)下的所有消息。
- 内部调用 `clearTopicMessagesThunk`
- **`createNewContext()`**:
- 发出一个全局事件 (`EVENT_NAMES.NEW_CONTEXT`),通常用于通知 UI 清空显示,准备新的上下文。不直接修改 Redux 状态。
- **`displayCount`**:
- (非操作函数) 从 Redux store 中获取当前的 `displayCount` 值。
- **`pauseMessages()`**:
- 尝试中止当前主题中正在进行的消息生成(状态为 `processing``pending`)。
- 通过查找相关的 `askId` 并调用 `abortCompletion` 来实现。
- 同时会 dispatch `setTopicLoading` action 将加载状态设为 `false`
- **`resumeMessage(message: Message, assistant: Assistant)`**:
- 恢复/重新发送一个用户消息。目前实现为直接调用 `resendMessage`
- **`regenerateAssistantMessage(message: Message, assistant: Assistant)`**:
- 重新生成指定的**助手**消息 (`message`) 的响应。
- 内部调用 `regenerateAssistantResponseThunk`
- **`appendAssistantResponse(existingAssistantMessage: Message, newModel: Model, assistant: Assistant)`**:
- 针对 `existingAssistantMessage` 所回复的**同一用户提问**,使用 `newModel` 追加一个新的助手响应。
- 内部调用 `appendAssistantResponseThunk`
- **`getTranslationUpdater(messageId: string, targetLanguage: string, sourceBlockId?: string, sourceLanguage?: string)`**:
- **用途**: 获取一个用于逐步更新翻译块内容的函数。
- **流程**:
1. 内部调用 `initiateTranslationThunk` 来创建或获取一个 `TRANSLATION` 类型的 `MessageBlock`,并获取其 `blockId`
2. 返回一个**异步更新函数**。
- **返回的更新函数 `(accumulatedText: string, isComplete?: boolean) => void`**:
- 接收累积的翻译文本和完成状态。
- 调用 `updateOneBlock` 更新 Redux 中的翻译块内容和状态 (`STREAMING``SUCCESS`)。
- 调用 `throttledBlockDbUpdate` 将更新(节流地)保存到数据库。
- 如果初始化失败Thunk 返回 `undefined`),则此函数返回 `null`
- **`createTopicBranch(sourceTopicId: string, branchPointIndex: number, newTopic: Topic)`**:
- 创建一个主题分支,将 `sourceTopicId` 主题中 `branchPointIndex` 索引之前的消息克隆到 `newTopic` 中。
- **注意**: `newTopic` 对象必须是调用此函数**之前**已经创建并添加到 Redux 和数据库中的。
- 内部调用 `cloneMessagesToNewTopicThunk`
## 依赖
- **`topic: Topic`**: 必须传入当前操作上下文的主题对象。Hook 返回的操作函数将始终作用于这个主题的 `topic.id`
- **Redux `dispatch`**: Hook 内部使用 `useAppDispatch` 获取 `dispatch` 函数来调用 actions 和 thunks。
## 相关 Hooks
在同一文件中还定义了两个辅助 Hook
- **`useTopicMessages(topic: Topic)`**:
- 使用 `selectMessagesForTopic` selector 来获取并返回指定主题的消息列表。
- **`useTopicLoading(topic: Topic)`**:
- 使用 `selectNewTopicLoading` selector 来获取并返回指定主题的加载状态。
这些 Hook 可以与 `useMessageOperations` 结合使用,方便地在组件中获取消息数据、加载状态,并执行相关操作。

View File

@@ -34,7 +34,7 @@
</a>
</h1>
<p align="center">
<a href="https://github.com/CherryHQ/cherry-studio">English</a> | 中文 | <a href="https://cherry-ai.com">官方网站</a> | <a href="https://docs.cherry-ai.com/cherry-studio-wen-dang/zh-cn">文档</a> | <a href="./dev.md">开发</a> | <a href="https://github.com/CherryHQ/cherry-studio/issues">反馈</a><br>
<a href="https://github.com/CherryHQ/cherry-studio">English</a> | 中文 | <a href="https://cherry-ai.com">官方网站</a> | <a href="https://docs.cherry-ai.com/cherry-studio-wen-dang/zh-cn">文档</a> | <a href="./guides/development.md">开发</a> | <a href="https://github.com/CherryHQ/cherry-studio/issues">反馈</a><br>
</p>
<!-- 题头徽章组合 -->
@@ -70,7 +70,7 @@ Cherry Studio 是一款支持多个大语言模型LLM服务商的桌面客
👏 欢迎加入 [Telegram 群组](https://t.me/CherryStudioAI)[Discord](https://discord.gg/wez8HtpxqQ) | [QQ群(575014769)](https://qm.qq.com/q/lo0D4qVZKi)
❤️ 喜欢 Cherry Studio? 点亮小星星 🌟 或 [赞助开发者](sponsor.md)! ❤️
❤️ 喜欢 Cherry Studio? 点亮小星星 🌟 或 [赞助开发者](./guides/sponsor.md)! ❤️
# 📖 使用教程
@@ -181,7 +181,7 @@ https://docs.cherry-ai.com
6. **社区参与**:加入讨论并帮助用户
7. **推广使用**:宣传 Cherry Studio
参考[分支策略](branching-strategy-zh.md)了解贡献指南
参考[分支策略](./guides/branching-strategy.md)了解贡献指南
## 入门
@@ -190,7 +190,7 @@ https://docs.cherry-ai.com
3. **提交更改**:提交并推送您的更改
4. **打开 Pull Request**:描述您的更改和原因
有关更详细的指南,请参阅我们的 [贡献指南](CONTRIBUTING.zh.md)
有关更详细的指南,请参阅我们的 [贡献指南](./guides/contributing.md)
感谢您的支持和贡献!

View File

@@ -16,7 +16,7 @@ Cherry Studio 采用结构化的分支策略来维护代码质量并简化开发
- 只接受文档更新和 bug 修复
- 经过完整测试后可以发布到生产环境
关于测试计划所使用的`testplan`分支,请查阅[测试计划](testplan-zh.md)。
关于测试计划所使用的`testplan`分支,请查阅[测试计划](./test-plan.md)。
## 贡献分支

View File

@@ -1,6 +1,6 @@
# Cherry Studio 贡献者指南
[**English**](../CONTRIBUTING.md) | [**中文**](CONTRIBUTING.zh.md)
[**English**](../../../CONTRIBUTING.md) | **中文**
欢迎来到 Cherry Studio 的贡献者社区!我们致力于将 Cherry Studio 打造成一个长期提供价值的项目,并希望邀请更多的开发者加入我们的行列。无论您是经验丰富的开发者还是刚刚起步的初学者,您的贡献都将帮助我们更好地服务用户,提升软件质量。
@@ -24,7 +24,7 @@
## 开始之前
请确保阅读了[行为准则](../CODE_OF_CONDUCT.md)和[LICENSE](../LICENSE)。
请确保阅读了[行为准则](../../../CODE_OF_CONDUCT.md)和[LICENSE](../../../LICENSE)。
## 开始贡献
@@ -32,7 +32,7 @@
### 测试
未经测试的功能等同于不存在。为确保代码真正有效,应通过单元测试和功能测试覆盖相关流程。因此,在考虑贡献时,也请考虑可测试性。所有测试均可本地运行,无需依赖 CI。请参阅[开发者指南](dev.md#test)中的Test部分。
未经测试的功能等同于不存在。为确保代码真正有效,应通过单元测试和功能测试覆盖相关流程。因此,在考虑贡献时,也请考虑可测试性。所有测试均可本地运行,无需依赖 CI。请参阅[开发者指南](./development.md#test)中的"Test"部分。
### 拉取请求的自动化测试
@@ -60,11 +60,11 @@ git commit --signoff -m "Your commit message"
### 获取代码审查/合并
维护者在此帮助您在合理时间内实现您的用例。他们会尽力在合理时间内审查您的代码并提供建设性反馈。但如果您在审查过程中受阻,或认为您的 Pull Request 未得到应有的关注,请通过 Issue 中的评论或者[社群](README.zh.md#-community)联系我们
维护者在此帮助您在合理时间内实现您的用例。他们会尽力在合理时间内审查您的代码并提供建设性反馈。但如果您在审查过程中受阻,或认为您的 Pull Request 未得到应有的关注,请通过 Issue 中的评论或者[社群](../README.md#-community)联系我们
### 参与测试计划
测试计划旨在为用户提供更稳定的应用体验和更快的迭代速度,详细情况请参阅[测试计划](testplan-zh.md)。
测试计划旨在为用户提供更稳定的应用体验和更快的迭代速度,详细情况请参阅[测试计划](./test-plan.md)。
### 其他建议

View File

@@ -0,0 +1,73 @@
# 🖥️ Develop
## IDE Setup
- Editor: [Cursor](https://www.cursor.com/), etc. Any VS Code compatible editor.
- Linter: [ESLint](https://marketplace.visualstudio.com/items?itemName=dbaeumer.vscode-eslint)
- Formatter: [Biome](https://marketplace.visualstudio.com/items?itemName=biomejs.biome)
## Project Setup
### Install
```bash
yarn
```
### Development
### Setup Node.js
Download and install [Node.js v22.x.x](https://nodejs.org/en/download)
### Setup Yarn
```bash
corepack enable
corepack prepare yarn@4.9.1 --activate
```
### Install Dependencies
```bash
yarn install
```
### ENV
```bash
copy .env.example .env
```
### Start
```bash
yarn dev
```
### Debug
```bash
yarn debug
```
Then input chrome://inspect in browser
### Test
```bash
yarn test
```
### Build
```bash
# For windows
$ yarn build:win
# For macOS
$ yarn build:mac
# For Linux
$ yarn build:linux
```

View File

@@ -15,11 +15,11 @@ i18n ally是一个强大的VSCode插件它能在开发阶段提供实时反
### 效果展示
![demo-1](./.assets.how-to-i18n/demo-1.png)
![demo-1](../../assets/images/i18n/demo-1.png)
![demo-2](./.assets.how-to-i18n/demo-2.png)
![demo-2](../../assets/images/i18n/demo-2.png)
![demo-3](./.assets.how-to-i18n/demo-3.png)
![demo-3](../../assets/images/i18n/demo-3.png)
## i18n 约定

View File

@@ -19,7 +19,7 @@
### 参与测试计划
开发者按照[贡献者指南](CONTRIBUTING.zh.md)要求正常提交`PR`并注意提交target为`main`)。仓库维护者会综合考虑(例如该功能对应用的影响程度,功能的重要性,是否需要更广泛的测试等),决定该`PR`是否应加入测试计划。
开发者按照[贡献者指南](./contributing.md)要求正常提交`PR`并注意提交target为`main`)。仓库维护者会综合考虑(例如该功能对应用的影响程度,功能的重要性,是否需要更广泛的测试等),决定该`PR`是否应加入测试计划。
若该`PR`加入测试计划,仓库维护者会做如下操作:

View File

@@ -85,7 +85,7 @@ graph TD
- **SvgPreview**: SVG 图像预览
- **GraphvizPreview**: Graphviz 图表预览
所有特殊视图组件共享通用架构,以确保一致的用户体验和功能。有关这些组件及其实现的详细信息,请参阅 [图像预览组件文档](./ImagePreview-zh.md)。
所有特殊视图组件共享通用架构,以确保一致的用户体验和功能。有关这些组件及其实现的详细信息,请参阅[图像预览组件文档](./image-preview.md)。
#### StatusBar 状态栏

View File

@@ -192,4 +192,4 @@ const { containerRef, error, isLoading, triggerRender, cancelRender, clearError,
- 共享状态管理
- 响应式布局适应
有关整体 CodeBlockView 架构的更多信息,请参阅 [CodeBlockView 文档](./CodeBlockView-zh.md)。
有关整体 CodeBlockView 架构的更多信息,请参阅 [CodeBlockView 文档](./code-block-view.md)。

View File

@@ -1,6 +1,24 @@
# `translate_languages` 表技术文档
# 数据库参考文档
## 📄 概述
本文档介绍 Cherry Studio 的数据库结构,包括设置字段和翻译语言表。
---
## 设置字段 (settings)
此部分包含设置相关字段的数据类型说明。
### 翻译相关字段
| 字段名 | 类型 | 说明 |
| ------------------------------ | ------------------------------ | ------------ |
| `translate:target:language` | `LanguageCode` | 翻译目标语言 |
| `translate:source:language` | `LanguageCode` | 翻译源语言 |
| `translate:bidirectional:pair` | `[LanguageCode, LanguageCode]` | 双向翻译对 |
---
## 翻译语言表 (translate_languages)
`translate_languages` 记录用户自定义的的语言类型(`Language`)。

View File

@@ -0,0 +1,404 @@
# 消息系统
本文档介绍 Cherry Studio 的消息系统架构,包括消息生命周期、状态管理和操作接口。
## 消息的生命周期
![消息生命周期](../../assets/images/message-lifecycle.png)
---
# messageBlock.ts 使用指南
该文件定义了用于管理应用程序中所有 `MessageBlock` 实体的 Redux Slice。它使用 Redux Toolkit 的 `createSlice``createEntityAdapter` 来高效地处理规范化的状态,并提供了一系列 actions 和 selectors 用于与消息块数据交互。
## 核心目标
- **状态管理**: 集中管理所有 `MessageBlock` 的状态。`MessageBlock` 代表消息中的不同内容单元(如文本、代码、图片、引用等)。
- **规范化**: 使用 `createEntityAdapter``MessageBlock` 数据存储在规范化的结构中(`{ ids: [], entities: {} }`),这有助于提高性能和简化更新逻辑。
- **可预测性**: 提供明确的 actions 来修改状态,并通过 selectors 安全地访问状态。
## 关键概念
- **Slice (`createSlice`)**: Redux Toolkit 的核心 API用于创建包含 reducer 逻辑、action creators 和初始状态的 Redux 模块。
- **Entity Adapter (`createEntityAdapter`)**: Redux Toolkit 提供的工具,用于简化对规范化数据的 CRUD创建、读取、更新、删除操作。它会自动生成 reducer 函数和 selectors。
- **Selectors**: 用于从 Redux store 中派生和计算数据的函数。Selectors 可以被记忆化memoized以提高性能。
## State 结构
`messageBlocks` slice 的状态结构由 `createEntityAdapter` 定义,大致如下:
```typescript
{
ids: string[]; // 存储所有 MessageBlock ID 的有序列表
entities: { [id: string]: MessageBlock }; // 按 ID 存储 MessageBlock 对象的字典
loadingState: 'idle' | 'loading' | 'succeeded' | 'failed'; // (可选) 其他状态,如加载状态
error: string | null; // (可选) 错误信息
}
```
## Actions
该 slice 导出以下 actions (由 `createSlice``createEntityAdapter` 自动生成或自定义)
- **`upsertOneBlock(payload: MessageBlock)`**:
- 添加一个新的 `MessageBlock` 或更新一个已存在的 `MessageBlock`。如果 payload 中的 `id` 已存在,则执行更新;否则执行插入。
- **`upsertManyBlocks(payload: MessageBlock[])`**:
- 添加或更新多个 `MessageBlock`。常用于批量加载数据(例如,加载一个 Topic 的所有消息块)。
- **`removeOneBlock(payload: string)`**:
- 根据提供的 `id` (payload) 移除单个 `MessageBlock`
- **`removeManyBlocks(payload: string[])`**:
- 根据提供的 `id` 数组 (payload) 移除多个 `MessageBlock`。常用于删除消息或清空 Topic 时清理相关的块。
- **`removeAllBlocks()`**:
- 移除 state 中的所有 `MessageBlock` 实体。
- **`updateOneBlock(payload: { id: string; changes: Partial<MessageBlock> })`**:
- 更新一个已存在的 `MessageBlock``payload` 需要包含块的 `id` 和一个包含要更改的字段的 `changes` 对象。
- **`setMessageBlocksLoading(payload: 'idle' | 'loading')`**:
- (自定义) 设置 `loadingState` 属性。
- **`setMessageBlocksError(payload: string)`**:
- (自定义) 设置 `loadingState``'failed'` 并记录错误信息。
**使用示例 (在 Thunk 或其他 Dispatch 的地方):**
```typescript
import { upsertOneBlock, removeManyBlocks, updateOneBlock } from './messageBlock'
import store from './store' // 假设这是你的 Redux store 实例
// 添加或更新一个块
const newBlock: MessageBlock = {
/* ... block data ... */
}
store.dispatch(upsertOneBlock(newBlock))
// 更新一个块的内容
store.dispatch(updateOneBlock({ id: blockId, changes: { content: 'New content' } }))
// 删除多个块
const blockIdsToRemove = ['id1', 'id2']
store.dispatch(removeManyBlocks(blockIdsToRemove))
```
## Selectors
该 slice 导出由 `createEntityAdapter` 生成的基础 selectors并通过 `messageBlocksSelectors` 对象访问:
- **`messageBlocksSelectors.selectIds(state: RootState): string[]`**: 返回包含所有块 ID 的数组。
- **`messageBlocksSelectors.selectEntities(state: RootState): { [id: string]: MessageBlock }`**: 返回块 ID 到块对象的映射字典。
- **`messageBlocksSelectors.selectAll(state: RootState): MessageBlock[]`**: 返回包含所有块对象的数组。
- **`messageBlocksSelectors.selectTotal(state: RootState): number`**: 返回块的总数。
- **`messageBlocksSelectors.selectById(state: RootState, id: string): MessageBlock | undefined`**: 根据 ID 返回单个块对象,如果找不到则返回 `undefined`
**此外,还提供了一个自定义的、记忆化的 selector**
- **`selectFormattedCitationsByBlockId(state: RootState, blockId: string | undefined): Citation[]`**:
- 接收一个 `blockId`
- 如果该 ID 对应的块是 `CITATION` 类型,则提取并格式化其包含的引用信息(来自网页搜索、知识库等),进行去重和重新编号,最后返回一个 `Citation[]` 数组,用于在 UI 中显示。
- 如果块不存在或类型不匹配,返回空数组 `[]`
- 这个 selector 封装了处理不同引用来源Gemini, OpenAI, OpenRouter, Zhipu 等)的复杂逻辑。
**使用示例 (在 React 组件或 `useSelector` 中):**
```typescript
import { useSelector } from 'react-redux'
import { messageBlocksSelectors, selectFormattedCitationsByBlockId } from './messageBlock'
import type { RootState } from './store'
// 获取所有块
const allBlocks = useSelector(messageBlocksSelectors.selectAll)
// 获取特定 ID 的块
const specificBlock = useSelector((state: RootState) => messageBlocksSelectors.selectById(state, someBlockId))
// 获取特定引用块格式化后的引用列表
const formattedCitations = useSelector((state: RootState) => selectFormattedCitationsByBlockId(state, citationBlockId))
// 在组件中使用引用数据
// {formattedCitations.map(citation => ...)}
```
## 集成
`messageBlock.ts` slice 通常与 `messageThunk.ts` 中的 Thunks 紧密协作。Thunks 负责处理异步逻辑(如 API 调用、数据库操作),并在需要时 dispatch `messageBlock` slice 的 actions 来更新状态。例如,当 `messageThunk` 接收到流式响应时,它会 dispatch `upsertOneBlock``updateOneBlock` 来实时更新对应的 `MessageBlock`。同样,删除消息的 Thunk 会 dispatch `removeManyBlocks`
理解 `messageBlock.ts` 的职责是管理**状态本身**,而 `messageThunk.ts` 负责**触发状态变更**的异步流程,这对于维护清晰的应用架构至关重要。
---
# messageThunk.ts 使用指南
该文件包含用于管理应用程序中消息流、处理助手交互以及同步 Redux 状态与 IndexedDB 数据库的核心 Thunk Action Creators。主要围绕 `Message``MessageBlock` 对象进行操作。
## 核心功能
1. **发送/接收消息**: 处理用户消息的发送,触发助手响应,并流式处理返回的数据,将其解析为不同的 `MessageBlock`
2. **状态管理**: 确保 Redux store 中的消息和消息块状态与 IndexedDB 中的持久化数据保持一致。
3. **消息操作**: 提供删除、重发、重新生成、编辑后重发、追加响应、克隆等消息生命周期管理功能。
4. **Block 处理**: 动态创建、更新和保存各种类型的 `MessageBlock`(文本、思考过程、工具调用、引用、图片、错误、翻译等)。
## 主要 Thunks
以下是一些关键的 Thunk 函数及其用途:
1. **`sendMessage(userMessage, userMessageBlocks, assistant, topicId)`**
- **用途**: 发送一条新的用户消息。
- **流程**:
- 保存用户消息 (`userMessage`) 及其块 (`userMessageBlocks`) 到 Redux 和 DB。
- 检查 `@mentions` 以确定是单模型响应还是多模型响应。
- 创建助手消息(们)的存根 (Stub)。
- 将存根添加到 Redux 和 DB。
- 将核心处理逻辑 `fetchAndProcessAssistantResponseImpl` 添加到该 `topicId` 的队列中以获取实际响应。
- **Block 相关**: 主要处理用户消息的初始 `MessageBlock` 保存。
2. **`fetchAndProcessAssistantResponseImpl(dispatch, getState, topicId, assistant, assistantMessage)`**
- **用途**: (内部函数) 获取并处理单个助手响应的核心逻辑,被 `sendMessage`, `resend...`, `regenerate...`, `append...` 等调用。
- **流程**:
- 设置 Topic 加载状态。
- 准备上下文消息。
- 调用 `fetchChatCompletion` API 服务。
- 使用 `createStreamProcessor` 处理流式响应。
- 通过各种回调 (`onTextChunk`, `onThinkingChunk`, `onToolCallComplete`, `onImageGenerated`, `onError`, `onComplete` 等) 处理不同类型的事件。
- **Block 相关**:
- 根据流事件创建初始 `UNKNOWN` 块。
- 实时创建和更新 `MAIN_TEXT``THINKING` 块,使用 `throttledBlockUpdate``throttledBlockDbUpdate` 进行节流更新。
- 创建 `TOOL`, `CITATION`, `IMAGE`, `ERROR` 等类型的块。
- 在事件完成时(如 `onTextComplete`, `onToolCallComplete`)将块状态标记为 `SUCCESS``ERROR`,并使用 `saveUpdatedBlockToDB` 保存最终状态。
- 使用 `handleBlockTransition` 管理非流式块(如 `TOOL`, `CITATION`)的添加和状态更新。
3. **`loadTopicMessagesThunk(topicId, forceReload)`**
- **用途**: 从数据库加载指定主题的所有消息及其关联的 `MessageBlock`
- **流程**:
- 从 DB 获取 `Topic` 及其 `messages` 列表。
- 根据消息 ID 列表从 DB 获取所有相关的 `MessageBlock`
- 使用 `upsertManyBlocks` 将块更新到 Redux。
- 将消息更新到 Redux。
- **Block 相关**: 负责将持久化的 `MessageBlock` 加载到 Redux 状态。
4. **删除 Thunks**
- `deleteSingleMessageThunk(topicId, messageId)`: 删除单个消息及其所有 `MessageBlock`
- `deleteMessageGroupThunk(topicId, askId)`: 删除一个用户消息及其所有相关的助手响应消息和它们的所有 `MessageBlock`
- `clearTopicMessagesThunk(topicId)`: 清空主题下的所有消息及其所有 `MessageBlock`
- **Block 相关**: 从 Redux 和 DB 中移除指定的 `MessageBlock`
5. **重发/重新生成 Thunks**
- `resendMessageThunk(topicId, userMessageToResend, assistant)`: 重发用户消息。会重置(清空 Block 并标记为 PENDING所有与该用户消息关联的助手响应然后重新请求生成。
- `resendUserMessageWithEditThunk(topicId, originalMessage, mainTextBlockId, editedContent, assistant)`: 用户编辑消息内容后重发。先更新用户消息的 `MAIN_TEXT` 块内容,然后调用 `resendMessageThunk`
- `regenerateAssistantResponseThunk(topicId, assistantMessageToRegenerate, assistant)`: 重新生成单个助手响应。重置该助手消息(清空 Block 并标记为 PENDING然后重新请求生成。
- **Block 相关**: 删除旧的 `MessageBlock`,并在重新生成过程中创建新的 `MessageBlock`
6. **`appendAssistantResponseThunk(topicId, existingAssistantMessageId, newModel, assistant)`**
- **用途**: 在已有的对话上下文中,针对同一个用户问题,使用新选择的模型追加一个新的助手响应。
- **流程**:
- 找到现有助手消息以获取原始 `askId`
- 创建使用 `newModel` 的新助手消息存根(使用相同的 `askId`)。
- 添加新存根到 Redux 和 DB。
-`fetchAndProcessAssistantResponseImpl` 添加到队列以生成新响应。
- **Block 相关**: 为新的助手响应创建全新的 `MessageBlock`
7. **`cloneMessagesToNewTopicThunk(sourceTopicId, branchPointIndex, newTopic)`**
- **用途**: 将源主题的部分消息(及其 Block克隆到一个**已存在**的新主题中。
- **流程**:
- 复制指定索引前的消息。
- 为所有克隆的消息和 Block 生成新的 UUID。
- 正确映射克隆消息之间的 `askId` 关系。
- 复制 `MessageBlock` 内容,更新其 `messageId` 指向新的消息 ID。
- 更新文件引用计数(如果 Block 是文件或图片)。
- 将克隆的消息和 Block 保存到新主题的 Redux 状态和 DB 中。
- **Block 相关**: 创建 `MessageBlock` 的副本,并更新其 ID 和 `messageId`
8. **`initiateTranslationThunk(messageId, topicId, targetLanguage, sourceBlockId?, sourceLanguage?)`**
- **用途**: 为指定消息启动翻译流程,创建一个初始的 `TRANSLATION` 类型的 `MessageBlock`
- **流程**:
- 创建一个状态为 `STREAMING``TranslationMessageBlock`
- 将其添加到 Redux 和 DB。
- 更新原消息的 `blocks` 列表以包含新的翻译块 ID。
- **Block 相关**: 创建并保存一个占位的 `TranslationMessageBlock`。实际翻译内容的获取和填充需要后续步骤。
## 内部机制和注意事项
- **数据库交互**: 通过 `saveMessageAndBlocksToDB`, `updateExistingMessageAndBlocksInDB`, `saveUpdatesToDB`, `saveUpdatedBlockToDB`, `throttledBlockDbUpdate` 等辅助函数与 IndexedDB (`db`) 交互,确保数据持久化。
- **状态同步**: Thunks 负责协调 Redux Store 和 IndexedDB 之间的数据一致性。
- **队列 (`getTopicQueue`)**: 使用 `AsyncQueue` 确保对同一主题的操作(尤其是 API 请求)按顺序执行,避免竞态条件。
- **节流 (`throttle`)**: 对流式响应中频繁的 Block 更新(文本、思考)使用 `lodash.throttle` 优化性能,减少 Redux dispatch 和 DB 写入次数。
- **错误处理**: `fetchAndProcessAssistantResponseImpl` 内的回调函数(特别是 `onError`)处理流处理和 API 调用中可能出现的错误,并创建 `ERROR` 类型的 `MessageBlock`
开发者在使用这些 Thunks 时,通常需要提供 `dispatch`, `getState` (由 Redux Thunk 中间件注入),以及如 `topicId`, `assistant` 配置对象, 相关的 `Message``MessageBlock` 对象/ID 等参数。理解每个 Thunk 的职责和它如何影响消息及块的状态至关重要。
---
# useMessageOperations.ts 使用指南
该文件定义了一个名为 `useMessageOperations` 的自定义 React Hook。这个 Hook 的主要目的是为 React 组件提供一个便捷的接口用于执行与特定主题Topic相关的各种消息操作。它封装了调用 Redux Thunks (`messageThunk.ts`) 和 Actions (`newMessage.ts`, `messageBlock.ts`) 的逻辑,简化了组件与消息数据交互的代码。
## 核心目标
- **封装**: 将复杂的消息操作逻辑(如删除、重发、重新生成、编辑、翻译等)封装在易于使用的函数中。
- **简化**: 让组件可以直接调用这些操作函数,而无需直接与 Redux `dispatch` 或 Thunks 交互。
- **上下文关联**: 所有操作都与传入的 `topic` 对象相关联,确保操作作用于正确的主题。
## 如何使用
在你的 React 函数组件中,导入并调用 `useMessageOperations` Hook并传入当前活动的 `Topic` 对象。
```typescript
import React from 'react';
import { useMessageOperations } from '@renderer/hooks/useMessageOperations';
import type { Topic, Message, Assistant, Model } from '@renderer/types';
interface MyComponentProps {
currentTopic: Topic;
currentAssistant: Assistant;
}
function MyComponent({ currentTopic, currentAssistant }: MyComponentProps) {
const {
deleteMessage,
resendMessage,
regenerateAssistantMessage,
appendAssistantResponse,
getTranslationUpdater,
createTopicBranch,
// ... 其他操作函数
} = useMessageOperations(currentTopic);
const handleDelete = (messageId: string) => {
deleteMessage(messageId);
};
const handleResend = (message: Message) => {
resendMessage(message, currentAssistant);
};
const handleAppend = (existingMsg: Message, newModel: Model) => {
appendAssistantResponse(existingMsg, newModel, currentAssistant);
}
// ... 在组件中使用其他操作函数
return (
<div>
{/* Component UI */}
<button onClick={() => handleDelete('some-message-id')}>Delete Message</button>
{/* ... */}
</div>
);
}
```
## 返回值
`useMessageOperations(topic)` Hook 返回一个包含以下函数和值的对象:
- **`deleteMessage(id: string)`**:
- 删除指定 `id` 的单个消息。
- 内部调用 `deleteSingleMessageThunk`
- **`deleteGroupMessages(askId: string)`**:
- 删除与指定 `askId` 相关联的一组消息(通常是用户提问及其所有助手回答)。
- 内部调用 `deleteMessageGroupThunk`
- **`editMessage(messageId: string, updates: Partial<Message>)`**:
- 更新指定 `messageId` 的消息的部分属性。
- **注意**: 目前主要用于更新 Redux 状态
- 内部调用 `newMessagesActions.updateMessage`
- **`resendMessage(message: Message, assistant: Assistant)`**:
- 重新发送指定的用户消息 (`message`),这将触发其所有关联助手响应的重新生成。
- 内部调用 `resendMessageThunk`
- **`resendUserMessageWithEdit(message: Message, editedContent: string, assistant: Assistant)`**:
- 在用户消息的主要文本块被编辑后,重新发送该消息。
- 会先查找消息的 `MAIN_TEXT` 块 ID然后调用 `resendUserMessageWithEditThunk`
- **`clearTopicMessages(_topicId?: string)`**:
- 清除当前主题(或可选的指定 `_topicId`)下的所有消息。
- 内部调用 `clearTopicMessagesThunk`
- **`createNewContext()`**:
- 发出一个全局事件 (`EVENT_NAMES.NEW_CONTEXT`),通常用于通知 UI 清空显示,准备新的上下文。不直接修改 Redux 状态。
- **`displayCount`**:
- (非操作函数) 从 Redux store 中获取当前的 `displayCount` 值。
- **`pauseMessages()`**:
- 尝试中止当前主题中正在进行的消息生成(状态为 `processing``pending`)。
- 通过查找相关的 `askId` 并调用 `abortCompletion` 来实现。
- 同时会 dispatch `setTopicLoading` action 将加载状态设为 `false`
- **`resumeMessage(message: Message, assistant: Assistant)`**:
- 恢复/重新发送一个用户消息。目前实现为直接调用 `resendMessage`
- **`regenerateAssistantMessage(message: Message, assistant: Assistant)`**:
- 重新生成指定的**助手**消息 (`message`) 的响应。
- 内部调用 `regenerateAssistantResponseThunk`
- **`appendAssistantResponse(existingAssistantMessage: Message, newModel: Model, assistant: Assistant)`**:
- 针对 `existingAssistantMessage` 所回复的**同一用户提问**,使用 `newModel` 追加一个新的助手响应。
- 内部调用 `appendAssistantResponseThunk`
- **`getTranslationUpdater(messageId: string, targetLanguage: string, sourceBlockId?: string, sourceLanguage?: string)`**:
- **用途**: 获取一个用于逐步更新翻译块内容的函数。
- **流程**:
1. 内部调用 `initiateTranslationThunk` 来创建或获取一个 `TRANSLATION` 类型的 `MessageBlock`,并获取其 `blockId`
2. 返回一个**异步更新函数**。
- **返回的更新函数 `(accumulatedText: string, isComplete?: boolean) => void`**:
- 接收累积的翻译文本和完成状态。
- 调用 `updateOneBlock` 更新 Redux 中的翻译块内容和状态 (`STREAMING``SUCCESS`)。
- 调用 `throttledBlockDbUpdate` 将更新(节流地)保存到数据库。
- 如果初始化失败Thunk 返回 `undefined`),则此函数返回 `null`
- **`createTopicBranch(sourceTopicId: string, branchPointIndex: number, newTopic: Topic)`**:
- 创建一个主题分支,将 `sourceTopicId` 主题中 `branchPointIndex` 索引之前的消息克隆到 `newTopic` 中。
- **注意**: `newTopic` 对象必须是调用此函数**之前**已经创建并添加到 Redux 和数据库中的。
- 内部调用 `cloneMessagesToNewTopicThunk`
## 依赖
- **`topic: Topic`**: 必须传入当前操作上下文的主题对象。Hook 返回的操作函数将始终作用于这个主题的 `topic.id`
- **Redux `dispatch`**: Hook 内部使用 `useAppDispatch` 获取 `dispatch` 函数来调用 actions 和 thunks。
## 相关 Hooks
在同一文件中还定义了两个辅助 Hook
- **`useTopicMessages(topic: Topic)`**:
- 使用 `selectMessagesForTopic` selector 来获取并返回指定主题的消息列表。
- **`useTopicLoading(topic: Topic)`**:
- 使用 `selectNewTopicLoading` selector 来获取并返回指定主题的加载状态。
这些 Hook 可以与 `useMessageOperations` 结合使用,方便地在组件中获取消息数据、加载状态,并执行相关操作。

View File

@@ -134,66 +134,108 @@ artifactBuildCompleted: scripts/artifact-build-completed.js
releaseInfo:
releaseNotes: |
<!--LANG:en-->
What's New in v1.7.0-rc.2
A New Era of Intelligence with Cherry Studio 1.7.1
✨ New Features:
- AI Models: Added support for Gemini 3, Gemini 3 Pro with image preview, and GPT-5.1
- Import: ChatGPT conversation import feature
- Agent: Git Bash detection and requirement check for Windows agents
- Search: Native language emoji search with CLDR data format
- Provider: Endpoint type support for cherryin provider
- Debug: Local crash mini dump file for better diagnostics
Today we're releasing Cherry Studio 1.7.1 — our most ambitious update yet, introducing Agent: autonomous AI that thinks, plans, and acts.
🐛 Important Bug Fixes:
- Error Handling: Improved error display in AiSdkToChunkAdapter
- Database: Optimized DatabaseManager and fixed libsql crash issues
- Memory: Fixed EventEmitter memory leak in useApiServer hook
- Messages: Fixed adjacent user messages appearing when assistant message contains error only
- Tools: Fixed missing execution state for approved tool permissions
- File Processing: Fixed "no such file" error for non-English filenames in open-mineru
- PDF: Fixed mineru PDF validation and 403 errors
- Images: Fixed base64 image save issues
- Search: Fixed URL context and web search capability
- Models: Added verbosity parameter support for GPT-5 models
- UI: Improved todo tool status icon visibility and colors
- Providers: Fixed api-host for vercel ai-gateway and gitcode update config
For years, AI assistants have been reactive — waiting for your commands, responding to your questions. With Agent, we're changing that. Now, AI can truly work alongside you: understanding complex goals, breaking them into steps, and executing them independently.
⚡ Improvements:
- SDK: Updated Google and OpenAI SDKs with new features
- UI: Simplified knowledge base creation modal and agent creation form
- Tools: Replaced renderToolContent function with ToolContent component
- Architecture: Namespace tool call IDs with session ID to prevent conflicts
- Config: AI SDK configuration refactoring
This is what we've been building toward. And it's just the beginning.
🤖 Meet Agent
Imagine having a brilliant colleague who never sleeps. Give Agent a goal — write a report, analyze data, refactor code — and watch it work. It reasons through problems, breaks them into steps, calls the right tools, and adapts when things change.
- **Think → Plan → Act**: From goal to execution, fully autonomous
- **Deep Reasoning**: Multi-turn thinking that solves real problems
- **Tool Mastery**: File operations, web search, code execution, and more
- **Skill Plugins**: Extend with custom commands and capabilities
- **You Stay in Control**: Real-time approval for sensitive actions
- **Full Visibility**: Every thought, every decision, fully transparent
🌐 Expanding Ecosystem
- **New Providers**: HuggingFace, Mistral, CherryIN, AI Gateway, Intel OVMS, Didi MCP
- **New Models**: Claude 4.5 Haiku, DeepSeek v3.2, GLM-4.6, Doubao, Ling series
- **MCP Integration**: Alibaba Cloud, ModelScope, Higress, MCP.so, TokenFlux and more
📚 Smarter Knowledge Base
- **OpenMinerU**: Self-hosted document processing
- **Full-Text Search**: Find anything instantly across your notes
- **Enhanced Tool Selection**: Smarter configuration for better AI assistance
📝 Notes, Reimagined
- Full-text search with highlighted results
- AI-powered smart rename
- Export as image
- Auto-wrap for tables
🖼️ Image & OCR
- Intel OVMS painting capabilities
- Intel OpenVINO NPU-accelerated OCR
🌍 Now in 10+ Languages
- Added German support
- Enhanced internationalization
⚡ Faster & More Polished
- Electron 38 upgrade
- New MCP management interface
- Dozens of UI refinements
❤️ Fully Open Source
Commercial restrictions removed. Cherry Studio now follows standard AGPL v3 — free for teams of any size.
The Agent Era is here. We can't wait to see what you'll create.
<!--LANG:zh-CN-->
v1.7.0-rc.2 新特性
Cherry Studio 1.7.1:开启智能新纪元
✨ 新功能:
- AI 模型:新增 Gemini 3、Gemini 3 Pro 图像预览支持,以及 GPT-5.1
- 导入ChatGPT 对话导入功能
- AgentWindows Agent 的 Git Bash 检测和要求检查
- 搜索:支持本地语言 emoji 搜索CLDR 数据格式)
- 提供商cherryin provider 的端点类型支持
- 调试:启用本地崩溃 mini dump 文件,方便诊断
今天,我们正式发布 Cherry Studio 1.7.1 —— 迄今最具雄心的版本,带来全新的 Agent能够自主思考、规划和行动的 AI。
🐛 重要修复:
- 错误处理:改进 AiSdkToChunkAdapter 的错误显示
- 数据库:优化 DatabaseManager 并修复 libsql 崩溃问题
- 内存:修复 useApiServer hook 中的 EventEmitter 内存泄漏
- 消息:修复当助手消息仅包含错误时相邻用户消息出现的问题
- 工具:修复批准工具权限缺少执行状态的问题
- 文件处理:修复 open-mineru 处理非英文文件名时的"无此文件"错误
- PDF修复 mineru PDF 验证和 403 错误
- 图片:修复 base64 图片保存问题
- 搜索:修复 URL 上下文和网络搜索功能
- 模型:为 GPT-5 模型添加 verbosity 参数支持
- UI改进 todo 工具状态图标可见性和颜色
- 提供商:修复 vercel ai-gateway 和 gitcode 更新配置的 api-host
多年来AI 助手一直是被动的——等待你的指令回应你的问题。Agent 改变了这一切。现在AI 能够真正与你并肩工作:理解复杂目标,将其拆解为步骤,并独立执行。
⚡ 改进:
- SDK更新 Google 和 OpenAI SDK新增功能和修复
- UI简化知识库创建模态框和 agent 创建表单
- 工具:用 ToolContent 组件替换 renderToolContent 函数,提升可读性
- 架构:用会话 ID 命名工具调用 ID 以防止冲突
- 配置AI SDK 配置重构
这是我们一直在构建的未来。而这,仅仅是开始。
🤖 认识 Agent
想象一位永不疲倦的得力伙伴。给 Agent 一个目标——撰写报告、分析数据、重构代码——然后看它工作。它会推理问题、拆解步骤、调用工具,并在情况变化时灵活应对。
- **思考 → 规划 → 行动**:从目标到执行,全程自主
- **深度推理**:多轮思考,解决真实问题
- **工具大师**:文件操作、网络搜索、代码执行,样样精通
- **技能插件**:自定义命令,无限扩展
- **你掌控全局**:敏感操作,实时审批
- **完全透明**:每一步思考,每一个决策,清晰可见
🌐 生态持续壮大
- **新增服务商**Hugging Face、Mistral、Perplexity、SophNet、AI Gateway、Cerebras AI
- **新增模型**Gemini 3、Gemini 3 Pro支持图像预览、GPT-5.1、Claude Opus 4.5
- **MCP 集成**百炼、魔搭、Higress、MCP.so、TokenFlux 等平台
📚 更智能的知识库
- **OpenMinerU**:本地自部署文档处理
- **全文搜索**:笔记内容一搜即达
- **增强工具选择**:更智能的配置,更好的 AI 协助
📝 笔记,焕然一新
- 全文搜索,结果高亮
- AI 智能重命名
- 导出为图片
- 表格自动换行
🖼️ 图像与 OCR
- Intel OVMS 绘图能力
- Intel OpenVINO NPU 加速 OCR
🌍 支持 10+ 种语言
- 新增德语支持
- 全面增强国际化
⚡ 更快、更精致
- 升级 Electron 38
- 新的 MCP 管理界面
- 数十处 UI 细节打磨
❤️ 完全开源
商用限制已移除。Cherry Studio 现遵循标准 AGPL v3 协议——任意规模团队均可自由使用。
Agent 纪元已至。期待你的创造。
<!--LANG:END-->

View File

@@ -25,7 +25,10 @@ export default defineConfig({
'@shared': resolve('packages/shared'),
'@logger': resolve('src/main/services/LoggerService'),
'@mcp-trace/trace-core': resolve('packages/mcp-trace/trace-core'),
'@mcp-trace/trace-node': resolve('packages/mcp-trace/trace-node')
'@mcp-trace/trace-node': resolve('packages/mcp-trace/trace-node'),
'@cherrystudio/ai-core/provider': resolve('packages/aiCore/src/core/providers'),
'@cherrystudio/ai-core': resolve('packages/aiCore/src'),
'@cherrystudio/ai-sdk-provider': resolve('packages/ai-sdk-provider/src')
}
},
build: {

View File

@@ -58,6 +58,7 @@ export default defineConfig([
'dist/**',
'out/**',
'local/**',
'tests/**',
'.yarn/**',
'.gitignore',
'scripts/cloudflare-worker.js',

View File

@@ -1,6 +1,6 @@
{
"name": "CherryStudio",
"version": "1.7.0-rc.1",
"version": "1.7.1",
"private": true,
"description": "A powerful AI assistant for producer.",
"main": "./out/main/index.js",
@@ -62,6 +62,7 @@
"test": "vitest run --silent",
"test:main": "vitest run --project main",
"test:renderer": "vitest run --project renderer",
"test:aicore": "vitest run --project aiCore",
"test:update": "yarn test:renderer --update",
"test:coverage": "vitest run --coverage --silent",
"test:ui": "vitest --ui",
@@ -80,7 +81,7 @@
"release:ai-sdk-provider": "yarn workspace @cherrystudio/ai-sdk-provider version patch --immediate && yarn workspace @cherrystudio/ai-sdk-provider build && yarn workspace @cherrystudio/ai-sdk-provider npm publish --access public"
},
"dependencies": {
"@anthropic-ai/claude-agent-sdk": "patch:@anthropic-ai/claude-agent-sdk@npm%3A0.1.30#~/.yarn/patches/@anthropic-ai-claude-agent-sdk-npm-0.1.30-b50a299674.patch",
"@anthropic-ai/claude-agent-sdk": "patch:@anthropic-ai/claude-agent-sdk@npm%3A0.1.53#~/.yarn/patches/@anthropic-ai-claude-agent-sdk-npm-0.1.53-4b77f4cf29.patch",
"@libsql/client": "0.14.0",
"@libsql/win32-x64-msvc": "^0.4.7",
"@napi-rs/system-ocr": "patch:@napi-rs/system-ocr@npm%3A1.0.2#~/.yarn/patches/@napi-rs-system-ocr-npm-1.0.2-59e7a78e8b.patch",
@@ -109,15 +110,15 @@
"@agentic/exa": "^7.3.3",
"@agentic/searxng": "^7.3.3",
"@agentic/tavily": "^7.3.3",
"@ai-sdk/amazon-bedrock": "^3.0.56",
"@ai-sdk/anthropic": "^2.0.45",
"@ai-sdk/amazon-bedrock": "^3.0.61",
"@ai-sdk/anthropic": "^2.0.49",
"@ai-sdk/cerebras": "^1.0.31",
"@ai-sdk/gateway": "^2.0.13",
"@ai-sdk/google": "patch:@ai-sdk/google@npm%3A2.0.40#~/.yarn/patches/@ai-sdk-google-npm-2.0.40-47e0eeee83.patch",
"@ai-sdk/google-vertex": "^3.0.72",
"@ai-sdk/gateway": "^2.0.15",
"@ai-sdk/google": "patch:@ai-sdk/google@npm%3A2.0.43#~/.yarn/patches/@ai-sdk-google-npm-2.0.43-689ed559b3.patch",
"@ai-sdk/google-vertex": "^3.0.79",
"@ai-sdk/huggingface": "^0.0.10",
"@ai-sdk/mistral": "^2.0.24",
"@ai-sdk/openai": "patch:@ai-sdk/openai@npm%3A2.0.71#~/.yarn/patches/@ai-sdk-openai-npm-2.0.71-a88ef00525.patch",
"@ai-sdk/openai": "patch:@ai-sdk/openai@npm%3A2.0.72#~/.yarn/patches/@ai-sdk-openai-npm-2.0.72-234e68da87.patch",
"@ai-sdk/perplexity": "^2.0.20",
"@ai-sdk/test-server": "^0.0.1",
"@ant-design/v5-patch-for-react-19": "^1.0.3",
@@ -164,15 +165,15 @@
"@modelcontextprotocol/sdk": "^1.17.5",
"@mozilla/readability": "^0.6.0",
"@notionhq/client": "^2.2.15",
"@openrouter/ai-sdk-provider": "^1.2.5",
"@openrouter/ai-sdk-provider": "^1.2.8",
"@opentelemetry/api": "^1.9.0",
"@opentelemetry/core": "2.0.0",
"@opentelemetry/exporter-trace-otlp-http": "^0.200.0",
"@opentelemetry/sdk-trace-base": "^2.0.0",
"@opentelemetry/sdk-trace-node": "^2.0.0",
"@opentelemetry/sdk-trace-web": "^2.0.0",
"@opeoginni/github-copilot-openai-compatible": "0.1.21",
"@playwright/test": "^1.52.0",
"@opeoginni/github-copilot-openai-compatible": "^0.1.21",
"@playwright/test": "^1.55.1",
"@radix-ui/react-context-menu": "^2.2.16",
"@reduxjs/toolkit": "^2.2.5",
"@shikijs/markdown-it": "^3.12.0",
@@ -217,8 +218,8 @@
"@types/mime-types": "^3",
"@types/node": "^22.17.1",
"@types/pako": "^1.0.2",
"@types/react": "^19.0.12",
"@types/react-dom": "^19.0.4",
"@types/react": "^19.2.7",
"@types/react-dom": "^19.2.3",
"@types/react-infinite-scroll-component": "^5.0.0",
"@types/react-transition-group": "^4.4.12",
"@types/react-window": "^1",
@@ -321,7 +322,6 @@
"p-queue": "^8.1.0",
"pdf-lib": "^1.17.1",
"pdf-parse": "^1.1.1",
"playwright": "^1.55.1",
"proxy-agent": "^6.5.0",
"react": "^19.2.0",
"react-dom": "^19.2.0",
@@ -412,12 +412,9 @@
"@langchain/openai@npm:>=0.1.0 <0.6.0": "patch:@langchain/openai@npm%3A1.0.0#~/.yarn/patches/@langchain-openai-npm-1.0.0-474d0ad9d4.patch",
"@langchain/openai@npm:^0.3.16": "patch:@langchain/openai@npm%3A1.0.0#~/.yarn/patches/@langchain-openai-npm-1.0.0-474d0ad9d4.patch",
"@langchain/openai@npm:>=0.2.0 <0.7.0": "patch:@langchain/openai@npm%3A1.0.0#~/.yarn/patches/@langchain-openai-npm-1.0.0-474d0ad9d4.patch",
"@ai-sdk/openai@npm:2.0.64": "patch:@ai-sdk/openai@npm%3A2.0.64#~/.yarn/patches/@ai-sdk-openai-npm-2.0.64-48f99f5bf3.patch",
"@ai-sdk/openai@npm:^2.0.42": "patch:@ai-sdk/openai@npm%3A2.0.71#~/.yarn/patches/@ai-sdk-openai-npm-2.0.71-a88ef00525.patch",
"@ai-sdk/google@npm:2.0.40": "patch:@ai-sdk/google@npm%3A2.0.40#~/.yarn/patches/@ai-sdk-google-npm-2.0.40-47e0eeee83.patch",
"@ai-sdk/openai@npm:2.0.71": "patch:@ai-sdk/openai@npm%3A2.0.71#~/.yarn/patches/@ai-sdk-openai-npm-2.0.71-a88ef00525.patch",
"@ai-sdk/openai-compatible@npm:1.0.27": "patch:@ai-sdk/openai-compatible@npm%3A1.0.27#~/.yarn/patches/@ai-sdk-openai-compatible-npm-1.0.27-06f74278cf.patch",
"@ai-sdk/openai-compatible@npm:^1.0.19": "patch:@ai-sdk/openai-compatible@npm%3A1.0.27#~/.yarn/patches/@ai-sdk-openai-compatible-npm-1.0.27-06f74278cf.patch"
"@ai-sdk/openai@npm:^2.0.42": "patch:@ai-sdk/openai@npm%3A2.0.72#~/.yarn/patches/@ai-sdk-openai-npm-2.0.72-234e68da87.patch",
"@ai-sdk/google@npm:^2.0.40": "patch:@ai-sdk/google@npm%3A2.0.40#~/.yarn/patches/@ai-sdk-google-npm-2.0.40-47e0eeee83.patch",
"@ai-sdk/openai-compatible@npm:^1.0.27": "patch:@ai-sdk/openai-compatible@npm%3A1.0.27#~/.yarn/patches/@ai-sdk-openai-compatible-npm-1.0.27-06f74278cf.patch"
},
"packageManager": "yarn@4.9.1",
"lint-staged": {

View File

@@ -69,6 +69,7 @@ export interface CherryInProviderSettings {
headers?: HeadersInput
/**
* Optional endpoint type to distinguish different endpoint behaviors.
* "image-generation" is also openai endpoint, but specifically for image generation.
*/
endpointType?: 'openai' | 'openai-response' | 'anthropic' | 'gemini' | 'image-generation' | 'jina-rerank'
}

View File

@@ -39,13 +39,13 @@
"ai": "^5.0.26"
},
"dependencies": {
"@ai-sdk/anthropic": "^2.0.45",
"@ai-sdk/azure": "^2.0.73",
"@ai-sdk/anthropic": "^2.0.49",
"@ai-sdk/azure": "^2.0.74",
"@ai-sdk/deepseek": "^1.0.29",
"@ai-sdk/openai-compatible": "patch:@ai-sdk/openai-compatible@npm%3A1.0.27#~/.yarn/patches/@ai-sdk-openai-compatible-npm-1.0.27-06f74278cf.patch",
"@ai-sdk/provider": "^2.0.0",
"@ai-sdk/provider-utils": "^3.0.17",
"@ai-sdk/xai": "^2.0.34",
"@ai-sdk/xai": "^2.0.36",
"zod": "^4.1.5"
},
"devDependencies": {

View File

@@ -3,12 +3,13 @@
* Provides realistic mock responses for all provider types
*/
import { jsonSchema, type ModelMessage, type Tool } from 'ai'
import type { ModelMessage, Tool } from 'ai'
import { jsonSchema } from 'ai'
/**
* Standard test messages for all scenarios
*/
export const testMessages = {
export const testMessages: Record<string, ModelMessage[]> = {
simple: [{ role: 'user' as const, content: 'Hello, how are you?' }],
conversation: [
@@ -45,7 +46,7 @@ export const testMessages = {
{ role: 'assistant' as const, content: '15 * 23 = 345' },
{ role: 'user' as const, content: 'Now divide that by 5' }
]
} satisfies Record<string, ModelMessage[]>
}
/**
* Standard test tools for tool calling scenarios
@@ -138,68 +139,17 @@ export const testTools: Record<string, Tool> = {
}
}
/**
* Mock streaming chunks for different providers
*/
export const mockStreamingChunks = {
text: [
{ type: 'text-delta' as const, textDelta: 'Hello' },
{ type: 'text-delta' as const, textDelta: ', ' },
{ type: 'text-delta' as const, textDelta: 'this ' },
{ type: 'text-delta' as const, textDelta: 'is ' },
{ type: 'text-delta' as const, textDelta: 'a ' },
{ type: 'text-delta' as const, textDelta: 'test.' }
],
withToolCall: [
{ type: 'text-delta' as const, textDelta: 'Let me check the weather for you.' },
{
type: 'tool-call-delta' as const,
toolCallType: 'function' as const,
toolCallId: 'call_123',
toolName: 'getWeather',
argsTextDelta: '{"location":'
},
{
type: 'tool-call-delta' as const,
toolCallType: 'function' as const,
toolCallId: 'call_123',
toolName: 'getWeather',
argsTextDelta: ' "San Francisco, CA"}'
},
{
type: 'tool-call' as const,
toolCallType: 'function' as const,
toolCallId: 'call_123',
toolName: 'getWeather',
args: { location: 'San Francisco, CA' }
}
],
withFinish: [
{ type: 'text-delta' as const, textDelta: 'Complete response.' },
{
type: 'finish' as const,
finishReason: 'stop' as const,
usage: {
promptTokens: 10,
completionTokens: 5,
totalTokens: 15
}
}
]
}
/**
* Mock complete responses for non-streaming scenarios
* Note: AI SDK v5 uses inputTokens/outputTokens instead of promptTokens/completionTokens
*/
export const mockCompleteResponses = {
simple: {
text: 'This is a simple response.',
finishReason: 'stop' as const,
usage: {
promptTokens: 15,
completionTokens: 8,
inputTokens: 15,
outputTokens: 8,
totalTokens: 23
}
},
@@ -215,8 +165,8 @@ export const mockCompleteResponses = {
],
finishReason: 'tool-calls' as const,
usage: {
promptTokens: 25,
completionTokens: 12,
inputTokens: 25,
outputTokens: 12,
totalTokens: 37
}
},
@@ -225,14 +175,15 @@ export const mockCompleteResponses = {
text: 'Response with warnings.',
finishReason: 'stop' as const,
usage: {
promptTokens: 10,
completionTokens: 5,
inputTokens: 10,
outputTokens: 5,
totalTokens: 15
},
warnings: [
{
type: 'unsupported-setting' as const,
message: 'Temperature parameter not supported for this model'
setting: 'temperature',
details: 'Temperature parameter not supported for this model'
}
]
}
@@ -285,47 +236,3 @@ export const mockImageResponses = {
warnings: []
}
}
/**
* Mock error responses
*/
export const mockErrors = {
invalidApiKey: {
name: 'APIError',
message: 'Invalid API key provided',
statusCode: 401
},
rateLimitExceeded: {
name: 'RateLimitError',
message: 'Rate limit exceeded. Please try again later.',
statusCode: 429,
headers: {
'retry-after': '60'
}
},
modelNotFound: {
name: 'ModelNotFoundError',
message: 'The requested model was not found',
statusCode: 404
},
contextLengthExceeded: {
name: 'ContextLengthError',
message: "This model's maximum context length is 4096 tokens",
statusCode: 400
},
timeout: {
name: 'TimeoutError',
message: 'Request timed out after 30000ms',
code: 'ETIMEDOUT'
},
networkError: {
name: 'NetworkError',
message: 'Network connection failed',
code: 'ECONNREFUSED'
}
}

View File

@@ -0,0 +1,35 @@
/**
* Mock for @cherrystudio/ai-sdk-provider
* This mock is used in tests to avoid importing the actual package
*/
export type CherryInProviderSettings = {
apiKey?: string
baseURL?: string
}
// oxlint-disable-next-line no-unused-vars
export const createCherryIn = (_options?: CherryInProviderSettings) => ({
// oxlint-disable-next-line no-unused-vars
languageModel: (_modelId: string) => ({
specificationVersion: 'v1',
provider: 'cherryin',
modelId: 'mock-model',
doGenerate: async () => ({ text: 'mock response' }),
doStream: async () => ({ stream: (async function* () {})() })
}),
// oxlint-disable-next-line no-unused-vars
chat: (_modelId: string) => ({
specificationVersion: 'v1',
provider: 'cherryin-chat',
modelId: 'mock-model',
doGenerate: async () => ({ text: 'mock response' }),
doStream: async () => ({ stream: (async function* () {})() })
}),
// oxlint-disable-next-line no-unused-vars
textEmbeddingModel: (_modelId: string) => ({
specificationVersion: 'v1',
provider: 'cherryin',
modelId: 'mock-embedding-model'
})
})

View File

@@ -0,0 +1,9 @@
/**
* Vitest Setup File
* Global test configuration and mocks for @cherrystudio/ai-core package
*/
// Mock Vite SSR helper to avoid Node environment errors
;(globalThis as any).__vite_ssr_exportName__ = (_name: string, value: any) => value
// Note: @cherrystudio/ai-sdk-provider is mocked via alias in vitest.config.ts

View File

@@ -0,0 +1,109 @@
import { describe, expect, it } from 'vitest'
import { createOpenAIOptions, createOpenRouterOptions, mergeProviderOptions } from '../factory'
describe('mergeProviderOptions', () => {
it('deep merges provider options for the same provider', () => {
const reasoningOptions = createOpenRouterOptions({
reasoning: {
enabled: true,
effort: 'medium'
}
})
const webSearchOptions = createOpenRouterOptions({
plugins: [{ id: 'web', max_results: 5 }]
})
const merged = mergeProviderOptions(reasoningOptions, webSearchOptions)
expect(merged.openrouter).toEqual({
reasoning: {
enabled: true,
effort: 'medium'
},
plugins: [{ id: 'web', max_results: 5 }]
})
})
it('preserves options from other providers while merging', () => {
const openRouter = createOpenRouterOptions({
reasoning: { enabled: true }
})
const openAI = createOpenAIOptions({
reasoningEffort: 'low'
})
const merged = mergeProviderOptions(openRouter, openAI)
expect(merged.openrouter).toEqual({ reasoning: { enabled: true } })
expect(merged.openai).toEqual({ reasoningEffort: 'low' })
})
it('overwrites primitive values with later values', () => {
const first = createOpenAIOptions({
reasoningEffort: 'low',
user: 'user-123'
})
const second = createOpenAIOptions({
reasoningEffort: 'high',
maxToolCalls: 5
})
const merged = mergeProviderOptions(first, second)
expect(merged.openai).toEqual({
reasoningEffort: 'high', // overwritten by second
user: 'user-123', // preserved from first
maxToolCalls: 5 // added from second
})
})
it('overwrites arrays with later values instead of merging', () => {
const first = createOpenRouterOptions({
models: ['gpt-4', 'gpt-3.5-turbo']
})
const second = createOpenRouterOptions({
models: ['claude-3-opus', 'claude-3-sonnet']
})
const merged = mergeProviderOptions(first, second)
// Array is completely replaced, not merged
expect(merged.openrouter?.models).toEqual(['claude-3-opus', 'claude-3-sonnet'])
})
it('deeply merges nested objects while overwriting primitives', () => {
const first = createOpenRouterOptions({
reasoning: {
enabled: true,
effort: 'low'
},
user: 'user-123'
})
const second = createOpenRouterOptions({
reasoning: {
effort: 'high',
max_tokens: 500
},
user: 'user-456'
})
const merged = mergeProviderOptions(first, second)
expect(merged.openrouter).toEqual({
reasoning: {
enabled: true, // preserved from first
effort: 'high', // overwritten by second
max_tokens: 500 // added from second
},
user: 'user-456' // overwritten by second
})
})
it('replaces arrays instead of merging them', () => {
const first = createOpenRouterOptions({ plugins: [{ id: 'old' }] })
const second = createOpenRouterOptions({ plugins: [{ id: 'new' }] })
const merged = mergeProviderOptions(first, second)
// @ts-expect-error type-check for openrouter options is skipped. see function signature of createOpenRouterOptions
expect(merged.openrouter?.plugins).toEqual([{ id: 'new' }])
})
})

View File

@@ -26,13 +26,65 @@ export function createGenericProviderOptions<T extends string>(
return { [provider]: options } as Record<T, Record<string, any>>
}
type PlainObject = Record<string, any>
const isPlainObject = (value: unknown): value is PlainObject => {
return typeof value === 'object' && value !== null && !Array.isArray(value)
}
function deepMergeObjects<T extends PlainObject>(target: T, source: PlainObject): T {
const result: PlainObject = { ...target }
Object.entries(source).forEach(([key, value]) => {
if (isPlainObject(value) && isPlainObject(result[key])) {
result[key] = deepMergeObjects(result[key], value)
} else {
result[key] = value
}
})
return result as T
}
/**
* 合并多个供应商的options
* @param optionsMap 包含多个供应商选项的对象
* @returns 合并后的TypedProviderOptions
* Deep-merge multiple provider-specific options.
* Nested objects are recursively merged; primitive values are overwritten.
*
* When the same key appears in multiple options:
* - If both values are plain objects: they are deeply merged (recursive merge)
* - If values are primitives/arrays: the later value overwrites the earlier one
*
* @example
* mergeProviderOptions(
* { openrouter: { reasoning: { enabled: true, effort: 'low' }, user: 'user-123' } },
* { openrouter: { reasoning: { effort: 'high', max_tokens: 500 }, models: ['gpt-4'] } }
* )
* // Result: {
* // openrouter: {
* // reasoning: { enabled: true, effort: 'high', max_tokens: 500 },
* // user: 'user-123',
* // models: ['gpt-4']
* // }
* // }
*
* @param optionsMap Objects containing options for multiple providers
* @returns Fully merged TypedProviderOptions
*/
export function mergeProviderOptions(...optionsMap: Partial<TypedProviderOptions>[]): TypedProviderOptions {
return Object.assign({}, ...optionsMap)
return optionsMap.reduce<TypedProviderOptions>((acc, options) => {
if (!options) {
return acc
}
Object.entries(options).forEach(([providerId, providerOptions]) => {
if (!providerOptions) {
return
}
if (acc[providerId]) {
acc[providerId] = deepMergeObjects(acc[providerId] as PlainObject, providerOptions as PlainObject)
} else {
acc[providerId] = providerOptions as any
}
})
return acc
}, {} as TypedProviderOptions)
}
/**

View File

@@ -19,15 +19,20 @@ describe('Provider Schemas', () => {
expect(Array.isArray(baseProviders)).toBe(true)
expect(baseProviders.length).toBeGreaterThan(0)
// These are the actual base providers defined in schemas.ts
const expectedIds = [
'openai',
'openai-responses',
'openai-chat',
'openai-compatible',
'anthropic',
'google',
'xai',
'azure',
'deepseek'
'azure-responses',
'deepseek',
'openrouter',
'cherryin',
'cherryin-chat'
]
const actualIds = baseProviders.map((p) => p.id)
expectedIds.forEach((id) => {

View File

@@ -232,11 +232,13 @@ describe('RuntimeExecutor.generateImage', () => {
expect(pluginCallOrder).toEqual(['onRequestStart', 'transformParams', 'transformResult', 'onRequestEnd'])
// transformParams receives params without model (model is handled separately)
// and context with core fields + dynamic fields (requestId, startTime, etc.)
expect(testPlugin.transformParams).toHaveBeenCalledWith(
{ prompt: 'A test image' },
expect.objectContaining({ prompt: 'A test image' }),
expect.objectContaining({
providerId: 'openai',
modelId: 'dall-e-3'
model: 'dall-e-3'
})
)
@@ -273,11 +275,12 @@ describe('RuntimeExecutor.generateImage', () => {
await executorWithPlugin.generateImage({ model: 'dall-e-3', prompt: 'A test image' })
// resolveModel receives model id and context with core fields
expect(modelResolutionPlugin.resolveModel).toHaveBeenCalledWith(
'dall-e-3',
expect.objectContaining({
providerId: 'openai',
modelId: 'dall-e-3'
model: 'dall-e-3'
})
)
@@ -339,12 +342,11 @@ describe('RuntimeExecutor.generateImage', () => {
.generateImage({ model: 'invalid-model', prompt: 'A test image' })
.catch((error) => error)
expect(thrownError).toBeInstanceOf(ImageGenerationError)
expect(thrownError.message).toContain('Failed to generate image:')
// Error is thrown from pluginEngine directly as ImageModelResolutionError
expect(thrownError).toBeInstanceOf(ImageModelResolutionError)
expect(thrownError.message).toContain('Failed to resolve image model: invalid-model')
expect(thrownError.providerId).toBe('openai')
expect(thrownError.modelId).toBe('invalid-model')
expect(thrownError.cause).toBeInstanceOf(ImageModelResolutionError)
expect(thrownError.cause.message).toContain('Failed to resolve image model: invalid-model')
})
it('should handle ImageModelResolutionError without provider', async () => {
@@ -362,8 +364,9 @@ describe('RuntimeExecutor.generateImage', () => {
const apiError = new Error('API request failed')
vi.mocked(aiGenerateImage).mockRejectedValue(apiError)
// Error propagates directly from pluginEngine without wrapping
await expect(executor.generateImage({ model: 'dall-e-3', prompt: 'A test image' })).rejects.toThrow(
'Failed to generate image:'
'API request failed'
)
})
@@ -376,8 +379,9 @@ describe('RuntimeExecutor.generateImage', () => {
vi.mocked(aiGenerateImage).mockRejectedValue(noImageError)
vi.mocked(NoImageGeneratedError.isInstance).mockReturnValue(true)
// Error propagates directly from pluginEngine
await expect(executor.generateImage({ model: 'dall-e-3', prompt: 'A test image' })).rejects.toThrow(
'Failed to generate image:'
'No image generated'
)
})
@@ -398,15 +402,17 @@ describe('RuntimeExecutor.generateImage', () => {
[errorPlugin]
)
// Error propagates directly from pluginEngine
await expect(executorWithPlugin.generateImage({ model: 'dall-e-3', prompt: 'A test image' })).rejects.toThrow(
'Failed to generate image:'
'Generation failed'
)
// onError receives the original error and context with core fields
expect(errorPlugin.onError).toHaveBeenCalledWith(
error,
expect.objectContaining({
providerId: 'openai',
modelId: 'dall-e-3'
model: 'dall-e-3'
})
)
})
@@ -419,9 +425,10 @@ describe('RuntimeExecutor.generateImage', () => {
const abortController = new AbortController()
setTimeout(() => abortController.abort(), 10)
// Error propagates directly from pluginEngine
await expect(
executor.generateImage({ model: 'dall-e-3', prompt: 'A test image', abortSignal: abortController.signal })
).rejects.toThrow('Failed to generate image:')
).rejects.toThrow('Operation was aborted')
})
})

View File

@@ -17,10 +17,14 @@ import type { AiPlugin } from '../../plugins'
import { globalRegistryManagement } from '../../providers/RegistryManagement'
import { RuntimeExecutor } from '../executor'
// Mock AI SDK
vi.mock('ai', () => ({
generateText: vi.fn()
}))
// Mock AI SDK - use importOriginal to keep jsonSchema and other non-mocked exports
vi.mock('ai', async (importOriginal) => {
const actual = (await importOriginal()) as Record<string, unknown>
return {
...actual,
generateText: vi.fn()
}
})
vi.mock('../../providers/RegistryManagement', () => ({
globalRegistryManagement: {
@@ -409,11 +413,12 @@ describe('RuntimeExecutor.generateText', () => {
})
).rejects.toThrow('Generation failed')
// onError receives the original error and context with core fields
expect(errorPlugin.onError).toHaveBeenCalledWith(
error,
expect.objectContaining({
providerId: 'openai',
modelId: 'gpt-4'
model: 'gpt-4'
})
)
})

View File

@@ -11,10 +11,14 @@ import type { AiPlugin } from '../../plugins'
import { globalRegistryManagement } from '../../providers/RegistryManagement'
import { RuntimeExecutor } from '../executor'
// Mock AI SDK
vi.mock('ai', () => ({
streamText: vi.fn()
}))
// Mock AI SDK - use importOriginal to keep jsonSchema and other non-mocked exports
vi.mock('ai', async (importOriginal) => {
const actual = (await importOriginal()) as Record<string, unknown>
return {
...actual,
streamText: vi.fn()
}
})
vi.mock('../../providers/RegistryManagement', () => ({
globalRegistryManagement: {
@@ -153,7 +157,7 @@ describe('RuntimeExecutor.streamText', () => {
describe('Max Tokens Parameter', () => {
const maxTokensValues = [10, 50, 100, 500, 1000, 2000, 4000]
it.each(maxTokensValues)('should support maxTokens=%s', async (maxTokens) => {
it.each(maxTokensValues)('should support maxOutputTokens=%s', async (maxOutputTokens) => {
const mockStream = {
textStream: (async function* () {
yield 'Response'
@@ -168,12 +172,13 @@ describe('RuntimeExecutor.streamText', () => {
await executor.streamText({
model: 'gpt-4',
messages: testMessages.simple,
maxOutputTokens: maxTokens
maxOutputTokens
})
// Parameters are passed through without transformation
expect(streamText).toHaveBeenCalledWith(
expect.objectContaining({
maxTokens
maxOutputTokens
})
)
})
@@ -513,11 +518,12 @@ describe('RuntimeExecutor.streamText', () => {
})
).rejects.toThrow('Stream error')
// onError receives the original error and context with core fields
expect(errorPlugin.onError).toHaveBeenCalledWith(
error,
expect.objectContaining({
providerId: 'openai',
modelId: 'gpt-4'
model: 'gpt-4'
})
)
})

View File

@@ -1,12 +1,20 @@
import path from 'node:path'
import { fileURLToPath } from 'node:url'
import { defineConfig } from 'vitest/config'
const __dirname = path.dirname(fileURLToPath(import.meta.url))
export default defineConfig({
test: {
globals: true
globals: true,
setupFiles: [path.resolve(__dirname, './src/__tests__/setup.ts')]
},
resolve: {
alias: {
'@': './src'
'@': path.resolve(__dirname, './src'),
// Mock external packages that may not be available in test environment
'@cherrystudio/ai-sdk-provider': path.resolve(__dirname, './src/__tests__/mocks/ai-sdk-provider.ts')
}
},
esbuild: {

View File

@@ -9,13 +9,27 @@
*/
import Anthropic from '@anthropic-ai/sdk'
import type { TextBlockParam } from '@anthropic-ai/sdk/resources'
import type { MessageCreateParams, TextBlockParam, Tool as AnthropicTool } from '@anthropic-ai/sdk/resources'
import { loggerService } from '@logger'
import type { Provider } from '@types'
import { type Provider, SystemProviderIds } from '@types'
import type { ModelMessage } from 'ai'
const logger = loggerService.withContext('anthropic-sdk')
/**
* Context for Anthropic SDK client creation.
* This allows the shared module to be used in different environments
* by providing environment-specific implementations.
*/
export interface AnthropicSdkContext {
/**
* Custom fetch function to use for HTTP requests.
* In Electron main process, this should be `net.fetch`.
* In other environments, can use the default fetch or a custom implementation.
*/
fetch?: typeof globalThis.fetch
}
const defaultClaudeCodeSystemPrompt = `You are Claude Code, Anthropic's official CLI for Claude.`
const defaultClaudeCodeSystem: Array<TextBlockParam> = [
@@ -58,8 +72,11 @@ const defaultClaudeCodeSystem: Array<TextBlockParam> = [
export function getSdkClient(
provider: Provider,
oauthToken?: string | null,
extraHeaders?: Record<string, string | string[]>
extraHeaders?: Record<string, string | string[]>,
context?: AnthropicSdkContext
): Anthropic {
const customFetch = context?.fetch
if (provider.authType === 'oauth') {
if (!oauthToken) {
throw new Error('OAuth token is not available')
@@ -85,14 +102,20 @@ export function getSdkClient(
'x-stainless-runtime': 'node',
'x-stainless-runtime-version': 'v22.18.0',
...extraHeaders
}
},
fetch: customFetch
})
}
const baseURL =
let baseURL =
provider.type === 'anthropic'
? provider.apiHost
: (provider.anthropicApiHost && provider.anthropicApiHost.trim()) || provider.apiHost
// Anthropic SDK automatically appends /v1 to all endpoints (like /v1/messages, /v1/models)
// We need to strip api version from baseURL to avoid duplication (e.g., /v3/v1/models)
// formatProviderApiHost adds /v1 for AI SDK compatibility, but Anthropic SDK needs it removed
baseURL = baseURL.replace(/\/v\d+(?:alpha|beta)?(?=\/|$)/i, '')
logger.debug('Anthropic API baseURL', { baseURL, providerId: provider.id })
if (provider.id === 'aihubmix') {
@@ -101,11 +124,12 @@ export function getSdkClient(
baseURL,
dangerouslyAllowBrowser: true,
defaultHeaders: {
'anthropic-beta': 'output-128k-2025-02-19',
'anthropic-beta': 'interleaved-thinking-2025-05-14',
'APP-Code': 'MLTG2087',
...provider.extra_headers,
...extraHeaders
}
},
fetch: customFetch
})
}
@@ -115,9 +139,11 @@ export function getSdkClient(
baseURL,
dangerouslyAllowBrowser: true,
defaultHeaders: {
'anthropic-beta': 'output-128k-2025-02-19',
'anthropic-beta': 'interleaved-thinking-2025-05-14',
Authorization: provider.id === SystemProviderIds.longcat ? `Bearer ${provider.apiKey}` : undefined,
...provider.extra_headers
}
},
fetch: customFetch
})
}
@@ -168,3 +194,31 @@ export function buildClaudeCodeSystemModelMessage(system?: string | Array<TextBl
content: block.text
}))
}
/**
* Sanitize tool definitions for Anthropic API.
*
* Removes non-standard fields like `input_examples` from tool definitions
* that Anthropic's API doesn't support. This prevents validation errors when
* tools with extended fields are passed to the Anthropic SDK.
*
* @param tools - Array of tool definitions from MessageCreateParams
* @returns Sanitized tools array with non-standard fields removed
*
* @example
* ```typescript
* const sanitizedTools = sanitizeToolsForAnthropic(request.tools)
* ```
*/
export function sanitizeToolsForAnthropic(tools?: MessageCreateParams['tools']): MessageCreateParams['tools'] {
if (!tools || tools.length === 0) return tools
return tools.map((tool) => {
if ('type' in tool && tool.type !== 'custom') return tool
// oxlint-disable-next-line no-unused-vars
const { input_examples, ...sanitizedTool } = tool as AnthropicTool & { input_examples?: unknown }
return sanitizedTool as typeof tool
})
}

View File

@@ -0,0 +1,245 @@
/**
* Shared API Utilities
*
* Common utilities for API URL formatting and validation.
* Used by both main process (API Server) and renderer.
*/
import type { MinimalProvider } from '@shared/provider'
import { trim } from 'lodash'
// Supported endpoints for routing
export const SUPPORTED_IMAGE_ENDPOINT_LIST = ['images/generations', 'images/edits', 'predict'] as const
export const SUPPORTED_ENDPOINT_LIST = [
'chat/completions',
'responses',
'messages',
'generateContent',
'streamGenerateContent',
...SUPPORTED_IMAGE_ENDPOINT_LIST
] as const
/**
* Removes the trailing slash from a URL string if it exists.
*/
export function withoutTrailingSlash<T extends string>(url: T): T {
return url.replace(/\/$/, '') as T
}
/**
* Matches a version segment in a path that starts with `/v<number>` and optionally
* continues with `alpha` or `beta`. The segment may be followed by `/` or the end
* of the string (useful for cases like `/v3alpha/resources`).
*/
const VERSION_REGEX_PATTERN = '\\/v\\d+(?:alpha|beta)?(?=\\/|$)'
/**
* Matches an API version at the end of a URL (with optional trailing slash).
* Used to detect and extract versions only from the trailing position.
*/
const TRAILING_VERSION_REGEX = /\/v\d+(?:alpha|beta)?\/?$/i
/**
* 判断 host 的 path 中是否包含形如版本的字符串(例如 /v1、/v2beta 等),
*
* @param host - 要检查的 host 或 path 字符串
* @returns 如果 path 中包含版本字符串则返回 true否则 false
*/
export function hasAPIVersion(host?: string): boolean {
if (!host) return false
const regex = new RegExp(VERSION_REGEX_PATTERN, 'i')
try {
const url = new URL(host)
return regex.test(url.pathname)
} catch {
// 若无法作为完整 URL 解析,则当作路径直接检测
return regex.test(host)
}
}
/**
* 格式化 Azure OpenAI 的 API 主机地址。
*/
export function formatAzureOpenAIApiHost(host: string): string {
const normalizedHost = withoutTrailingSlash(host)
?.replace(/\/v1$/, '')
.replace(/\/openai$/, '')
// NOTE: AISDK会添加上`v1`
return formatApiHost(normalizedHost + '/openai', false)
}
export function formatVertexApiHost(
provider: MinimalProvider,
project: string = 'test-project',
location: string = 'us-central1'
): string {
const { apiHost } = provider
const trimmedHost = withoutTrailingSlash(trim(apiHost))
if (!trimmedHost || trimmedHost.endsWith('aiplatform.googleapis.com')) {
const host =
location === 'global' ? 'https://aiplatform.googleapis.com' : `https://${location}-aiplatform.googleapis.com`
return `${formatApiHost(host)}/projects/${project}/locations/${location}`
}
return formatApiHost(trimmedHost)
}
/**
* Formats an API host URL by normalizing it and optionally appending an API version.
*
* @param host - The API host URL to format. Leading/trailing whitespace will be trimmed and trailing slashes removed.
* @param supportApiVersion - Whether the API version is supported. Defaults to `true`.
* @param apiVersion - The API version to append if needed. Defaults to `'v1'`.
*
* @returns The formatted API host URL. If the host is empty after normalization, returns an empty string.
* If the host ends with '#', API version is not supported, or the host already contains a version, returns the normalized host as-is.
* Otherwise, returns the host with the API version appended.
*
* @example
* formatApiHost('https://api.example.com/') // Returns 'https://api.example.com/v1'
* formatApiHost('https://api.example.com#') // Returns 'https://api.example.com#'
* formatApiHost('https://api.example.com/v2', true, 'v1') // Returns 'https://api.example.com/v2'
*/
export function formatApiHost(host?: string, supportApiVersion: boolean = true, apiVersion: string = 'v1'): string {
const normalizedHost = withoutTrailingSlash(trim(host))
if (!normalizedHost) {
return ''
}
if (normalizedHost.endsWith('#') || !supportApiVersion || hasAPIVersion(normalizedHost)) {
return normalizedHost
}
return `${normalizedHost}/${apiVersion}`
}
/**
* Converts an API host URL into separate base URL and endpoint components.
*
* This function extracts endpoint information from a composite API host string.
* If the host ends with '#', it attempts to match the preceding part against the supported endpoint list.
*
* @param apiHost - The API host string to parse
* @returns An object containing:
* - `baseURL`: The base URL without the endpoint suffix
* - `endpoint`: The matched endpoint identifier, or empty string if no match found
*
* @example
* routeToEndpoint('https://api.example.com/openai/chat/completions#')
* // Returns: { baseURL: 'https://api.example.com/v1', endpoint: 'chat/completions' }
*
* @example
* routeToEndpoint('https://api.example.com/v1')
* // Returns: { baseURL: 'https://api.example.com/v1', endpoint: '' }
*/
export function routeToEndpoint(apiHost: string): { baseURL: string; endpoint: string } {
const trimmedHost = (apiHost || '').trim()
if (!trimmedHost.endsWith('#')) {
return { baseURL: trimmedHost, endpoint: '' }
}
// Remove trailing #
const host = trimmedHost.slice(0, -1)
const endpointMatch = SUPPORTED_ENDPOINT_LIST.find((endpoint) => host.endsWith(endpoint))
if (!endpointMatch) {
const baseURL = withoutTrailingSlash(host)
return { baseURL, endpoint: '' }
}
const baseSegment = host.slice(0, host.length - endpointMatch.length)
const baseURL = withoutTrailingSlash(baseSegment).replace(/:$/, '') // Remove trailing colon (gemini special case)
return { baseURL, endpoint: endpointMatch }
}
/**
* Gets the AI SDK compatible base URL from a provider's apiHost.
*
* AI SDK expects baseURL WITH version suffix (e.g., /v1).
* This function:
* 1. Handles '#' endpoint routing format
* 2. Ensures the URL has a version suffix (adds /v1 if missing)
*
* @param apiHost - The provider's apiHost value (may or may not have /v1)
* @param apiVersion - The API version to use if missing. Defaults to 'v1'.
* @returns The baseURL suitable for AI SDK (with version suffix)
*
* @example
* getAiSdkBaseUrl('https://api.openai.com') // 'https://api.openai.com/v1'
* getAiSdkBaseUrl('https://api.openai.com/v1') // 'https://api.openai.com/v1'
* getAiSdkBaseUrl('https://api.example.com/chat/completions#') // 'https://api.example.com'
*/
export function getAiSdkBaseUrl(apiHost: string, apiVersion: string = 'v1'): string {
// First handle '#' endpoint routing format
const { baseURL } = routeToEndpoint(apiHost)
// If already has version, return as-is
if (hasAPIVersion(baseURL)) {
return withoutTrailingSlash(baseURL)
}
// Add version suffix
return `${withoutTrailingSlash(baseURL)}/${apiVersion}`
}
/**
* Validates an API host address.
*
* @param apiHost - The API host address to validate
* @returns true if valid URL with http/https protocol, false otherwise
*/
export function validateApiHost(apiHost: string): boolean {
if (!apiHost || !apiHost.trim()) {
return true // Allow empty
}
try {
const url = new URL(apiHost.trim())
return url.protocol === 'http:' || url.protocol === 'https:'
} catch {
return false
}
}
/**
* Extracts the trailing API version segment from a URL path.
*
* This function extracts API version patterns (e.g., `v1`, `v2beta`) from the end of a URL.
* Only versions at the end of the path are extracted, not versions in the middle.
* The returned version string does not include leading or trailing slashes.
*
* @param {string} url - The URL string to parse.
* @returns {string | undefined} The trailing API version found (e.g., 'v1', 'v2beta'), or undefined if none found.
*
* @example
* getTrailingApiVersion('https://api.example.com/v1') // 'v1'
* getTrailingApiVersion('https://api.example.com/v2beta/') // 'v2beta'
* getTrailingApiVersion('https://api.example.com/v1/chat') // undefined (version not at end)
* getTrailingApiVersion('https://gateway.ai.cloudflare.com/v1/xxx/v1beta') // 'v1beta'
* getTrailingApiVersion('https://api.example.com') // undefined
*/
export function getTrailingApiVersion(url: string): string | undefined {
const match = url.match(TRAILING_VERSION_REGEX)
if (match) {
// Extract version without leading slash and trailing slash
return match[0].replace(/^\//, '').replace(/\/$/, '')
}
return undefined
}
/**
* Removes the trailing API version segment from a URL path.
*
* This function removes API version patterns (e.g., `/v1`, `/v2beta`) from the end of a URL.
* Only versions at the end of the path are removed, not versions in the middle.
*
* @param {string} url - The URL string to process.
* @returns {string} The URL with the trailing API version removed, or the original URL if no trailing version found.
*
* @example
* withoutTrailingApiVersion('https://api.example.com/v1') // 'https://api.example.com'
* withoutTrailingApiVersion('https://api.example.com/v2beta/') // 'https://api.example.com'
* withoutTrailingApiVersion('https://api.example.com/v1/chat') // 'https://api.example.com/v1/chat' (no change)
* withoutTrailingApiVersion('https://api.example.com') // 'https://api.example.com'
*/
export function withoutTrailingApiVersion(url: string): string {
return url.replace(TRAILING_VERSION_REGEX, '')
}

View File

@@ -0,0 +1,77 @@
/**
* @fileoverview Shared provider configuration for Claude Code and Anthropic API compatibility
*
* This module defines which models from specific providers support the Anthropic API endpoint.
* Used by both the Code Tools page and the Anthropic SDK client.
*/
/**
* Silicon provider models that support Anthropic API endpoint.
* These models can be used with Claude Code via the Anthropic-compatible API.
*
* @see https://docs.siliconflow.cn/cn/api-reference/chat-completions/messages
*/
export const SILICON_ANTHROPIC_COMPATIBLE_MODELS: readonly string[] = [
// DeepSeek V3.1 series
'Pro/deepseek-ai/DeepSeek-V3.1-Terminus',
'deepseek-ai/DeepSeek-V3.1',
'Pro/deepseek-ai/DeepSeek-V3.1',
// DeepSeek V3 series
'deepseek-ai/DeepSeek-V3',
'Pro/deepseek-ai/DeepSeek-V3',
// Moonshot/Kimi series
'moonshotai/Kimi-K2-Instruct-0905',
'Pro/moonshotai/Kimi-K2-Instruct-0905',
'moonshotai/Kimi-Dev-72B',
// Baidu ERNIE
'baidu/ERNIE-4.5-300B-A47B'
]
/**
* Creates a Set for efficient lookup of silicon Anthropic-compatible model IDs.
*/
const SILICON_ANTHROPIC_COMPATIBLE_MODEL_SET = new Set(SILICON_ANTHROPIC_COMPATIBLE_MODELS)
/**
* Checks if a model ID is compatible with Anthropic API on Silicon provider.
*
* @param modelId - The model ID to check
* @returns true if the model supports Anthropic API endpoint
*/
export function isSiliconAnthropicCompatibleModel(modelId: string): boolean {
return SILICON_ANTHROPIC_COMPATIBLE_MODEL_SET.has(modelId)
}
/**
* PPIO provider models that support Anthropic API endpoint.
* These models can be used with Claude Code via the Anthropic-compatible API.
*
* @see https://ppio.com/docs/model/llm-anthropic-compatibility
*/
export const PPIO_ANTHROPIC_COMPATIBLE_MODELS: readonly string[] = [
'moonshotai/kimi-k2-thinking',
'minimax/minimax-m2',
'deepseek/deepseek-v3.2-exp',
'deepseek/deepseek-v3.1-terminus',
'zai-org/glm-4.6',
'moonshotai/kimi-k2-0905',
'deepseek/deepseek-v3.1',
'moonshotai/kimi-k2-instruct',
'qwen/qwen3-next-80b-a3b-instruct',
'qwen/qwen3-next-80b-a3b-thinking'
]
/**
* Creates a Set for efficient lookup of PPIO Anthropic-compatible model IDs.
*/
const PPIO_ANTHROPIC_COMPATIBLE_MODEL_SET = new Set(PPIO_ANTHROPIC_COMPATIBLE_MODELS)
/**
* Checks if a model ID is compatible with Anthropic API on PPIO provider.
*
* @param modelId - The model ID to check
* @returns true if the model supports Anthropic API endpoint
*/
export function isPpioAnthropicCompatibleModel(modelId: string): boolean {
return PPIO_ANTHROPIC_COMPATIBLE_MODEL_SET.has(modelId)
}

View File

@@ -0,0 +1,15 @@
/**
* Shared AI SDK Middlewares
*
* Environment-agnostic middlewares that can be used in both
* renderer process and main process (API server).
*/
export {
buildSharedMiddlewares,
getReasoningTagName,
isGemini3ModelId,
openrouterReasoningMiddleware,
type SharedMiddlewareConfig,
skipGeminiThoughtSignatureMiddleware
} from './middlewares'

View File

@@ -0,0 +1,205 @@
/**
* Shared AI SDK Middlewares
*
* These middlewares are environment-agnostic and can be used in both
* renderer process and main process (API server).
*/
import type { LanguageModelV2Middleware, LanguageModelV2StreamPart } from '@ai-sdk/provider'
import { extractReasoningMiddleware } from 'ai'
/**
* Configuration for building shared middlewares
*/
export interface SharedMiddlewareConfig {
/**
* Whether to enable reasoning extraction
*/
enableReasoning?: boolean
/**
* Tag name for reasoning extraction
* Defaults based on model ID
*/
reasoningTagName?: string
/**
* Model ID - used to determine default reasoning tag and model detection
*/
modelId?: string
/**
* Provider ID (Cherry Studio provider ID)
* Used for provider-specific middlewares like OpenRouter
*/
providerId?: string
/**
* AI SDK Provider ID
* Used for Gemini thought signature middleware
* e.g., 'google', 'google-vertex'
*/
aiSdkProviderId?: string
}
/**
* Check if model ID represents a Gemini 3 (2.5) model
* that requires thought signature handling
*
* @param modelId - The model ID string (not Model object)
*/
export function isGemini3ModelId(modelId?: string): boolean {
if (!modelId) return false
const lowerModelId = modelId.toLowerCase()
return lowerModelId.includes('gemini-3')
}
/**
* Get the default reasoning tag name based on model ID
*
* Different models use different tags for reasoning content:
* - Most models: 'think'
* - GPT-OSS models: 'reasoning'
* - Gemini models: 'thought'
* - Seed models: 'seed:think'
*/
export function getReasoningTagName(modelId?: string): string {
if (!modelId) return 'think'
const lowerModelId = modelId.toLowerCase()
if (lowerModelId.includes('gpt-oss')) return 'reasoning'
if (lowerModelId.includes('gemini')) return 'thought'
if (lowerModelId.includes('seed-oss-36b')) return 'seed:think'
return 'think'
}
/**
* Skip Gemini Thought Signature Middleware
*
* Due to the complexity of multi-model client requests (which can switch
* to other models mid-process), this middleware skips all Gemini 3
* thinking signatures validation.
*
* @param aiSdkId - AI SDK Provider ID (e.g., 'google', 'google-vertex')
* @returns LanguageModelV2Middleware
*/
export function skipGeminiThoughtSignatureMiddleware(aiSdkId: string): LanguageModelV2Middleware {
const MAGIC_STRING = 'skip_thought_signature_validator'
return {
middlewareVersion: 'v2',
transformParams: async ({ params }) => {
const transformedParams = { ...params }
// Process messages in prompt
if (transformedParams.prompt && Array.isArray(transformedParams.prompt)) {
transformedParams.prompt = transformedParams.prompt.map((message) => {
if (typeof message.content !== 'string') {
for (const part of message.content) {
const googleOptions = part?.providerOptions?.[aiSdkId]
if (googleOptions?.thoughtSignature) {
googleOptions.thoughtSignature = MAGIC_STRING
}
}
}
return message
})
}
return transformedParams
}
}
}
/**
* OpenRouter Reasoning Middleware
*
* Filters out [REDACTED] blocks from OpenRouter reasoning responses.
* OpenRouter may include [REDACTED] markers in reasoning content that
* should be removed for cleaner output.
*
* @see https://openrouter.ai/docs/docs/best-practices/reasoning-tokens
* @returns LanguageModelV2Middleware
*/
export function openrouterReasoningMiddleware(): LanguageModelV2Middleware {
const REDACTED_BLOCK = '[REDACTED]'
return {
middlewareVersion: 'v2',
wrapGenerate: async ({ doGenerate }) => {
const { content, ...rest } = await doGenerate()
const modifiedContent = content.map((part) => {
if (part.type === 'reasoning' && part.text.includes(REDACTED_BLOCK)) {
return {
...part,
text: part.text.replace(REDACTED_BLOCK, '')
}
}
return part
})
return { content: modifiedContent, ...rest }
},
wrapStream: async ({ doStream }) => {
const { stream, ...rest } = await doStream()
return {
stream: stream.pipeThrough(
new TransformStream<LanguageModelV2StreamPart, LanguageModelV2StreamPart>({
transform(
chunk: LanguageModelV2StreamPart,
controller: TransformStreamDefaultController<LanguageModelV2StreamPart>
) {
if (chunk.type === 'reasoning-delta' && chunk.delta.includes(REDACTED_BLOCK)) {
controller.enqueue({
...chunk,
delta: chunk.delta.replace(REDACTED_BLOCK, '')
})
} else {
controller.enqueue(chunk)
}
}
})
),
...rest
}
}
}
}
/**
* Build shared middlewares based on configuration
*
* This function builds a set of middlewares that are commonly needed
* across different environments (renderer, API server).
*
* @param config - Configuration for middleware building
* @returns Array of AI SDK middlewares
*
* @example
* ```typescript
* import { buildSharedMiddlewares } from '@shared/middleware'
*
* const middlewares = buildSharedMiddlewares({
* enableReasoning: true,
* modelId: 'gemini-2.5-pro',
* providerId: 'openrouter',
* aiSdkProviderId: 'google'
* })
* ```
*/
export function buildSharedMiddlewares(config: SharedMiddlewareConfig): LanguageModelV2Middleware[] {
const middlewares: LanguageModelV2Middleware[] = []
// 1. Reasoning extraction middleware
if (config.enableReasoning) {
const tagName = config.reasoningTagName || getReasoningTagName(config.modelId)
middlewares.push(extractReasoningMiddleware({ tagName }))
}
// 2. OpenRouter-specific: filter [REDACTED] blocks
if (config.providerId === 'openrouter' && config.enableReasoning) {
middlewares.push(openrouterReasoningMiddleware())
}
// 3. Gemini 3 (2.5) specific: skip thought signature validation
if (isGemini3ModelId(config.modelId) && config.aiSdkProviderId) {
middlewares.push(skipGeminiThoughtSignatureMiddleware(config.aiSdkProviderId))
}
return middlewares
}

View File

@@ -1,13 +1,13 @@
/**
* AiHubMix规则集
*/
import { isOpenAILLMModel } from '@renderer/config/models'
import type { Provider } from '@renderer/types'
import { getLowerBaseModelName } from '@shared/utils/naming'
import type { MinimalModel, MinimalProvider } from '../types'
import { provider2Provider, startsWith } from './helper'
import type { RuleSet } from './types'
const extraProviderConfig = (provider: Provider) => {
const extraProviderConfig = <P extends MinimalProvider>(provider: P) => {
return {
...provider,
extra_headers: {
@@ -17,11 +17,23 @@ const extraProviderConfig = (provider: Provider) => {
}
}
function isOpenAILLMModel<M extends MinimalModel>(model: M): boolean {
const modelId = getLowerBaseModelName(model.id)
const reasonings = ['o1', 'o3', 'o4', 'gpt-oss']
if (reasonings.some((r) => modelId.includes(r))) {
return true
}
if (modelId.includes('gpt')) {
return true
}
return false
}
const AIHUBMIX_RULES: RuleSet = {
rules: [
{
match: startsWith('claude'),
provider: (provider: Provider) => {
provider: (provider) => {
return extraProviderConfig({
...provider,
type: 'anthropic'
@@ -34,7 +46,7 @@ const AIHUBMIX_RULES: RuleSet = {
!model.id.endsWith('-nothink') &&
!model.id.endsWith('-search') &&
!model.id.includes('embedding'),
provider: (provider: Provider) => {
provider: (provider) => {
return extraProviderConfig({
...provider,
type: 'gemini',
@@ -44,7 +56,7 @@ const AIHUBMIX_RULES: RuleSet = {
},
{
match: isOpenAILLMModel,
provider: (provider: Provider) => {
provider: (provider) => {
return extraProviderConfig({
...provider,
type: 'openai-response'
@@ -52,7 +64,8 @@ const AIHUBMIX_RULES: RuleSet = {
}
}
],
fallbackRule: (provider: Provider) => extraProviderConfig(provider)
fallbackRule: (provider) => extraProviderConfig(provider)
}
export const aihubmixProviderCreator = provider2Provider.bind(null, AIHUBMIX_RULES)
export const aihubmixProviderCreator = <P extends MinimalProvider>(model: MinimalModel, provider: P): P =>
provider2Provider<MinimalModel, MinimalProvider, P>(AIHUBMIX_RULES, model, provider)

View File

@@ -0,0 +1,22 @@
import type { MinimalModel, MinimalProvider, ProviderType } from '../types'
import { provider2Provider, startsWith } from './helper'
import type { RuleSet } from './types'
// https://platform.claude.com/docs/en/build-with-claude/claude-in-microsoft-foundry
const AZURE_ANTHROPIC_RULES: RuleSet = {
rules: [
{
match: startsWith('claude'),
provider: (provider: MinimalProvider) => ({
...provider,
type: 'anthropic' as ProviderType,
apiHost: provider.apiHost + 'anthropic/v1',
id: 'azure-anthropic'
})
}
],
fallbackRule: (provider: MinimalProvider) => provider
}
export const azureAnthropicProviderCreator = <P extends MinimalProvider>(model: MinimalModel, provider: P): P =>
provider2Provider<MinimalModel, MinimalProvider, P>(AZURE_ANTHROPIC_RULES, model, provider)

View File

@@ -0,0 +1,32 @@
import type { MinimalModel, MinimalProvider } from '../types'
import type { RuleSet } from './types'
export const startsWith =
(prefix: string) =>
<M extends MinimalModel>(model: M) =>
model.id.toLowerCase().startsWith(prefix.toLowerCase())
export const endpointIs =
(type: string) =>
<M extends MinimalModel>(model: M) =>
model.endpoint_type === type
/**
* 解析模型对应的Provider
* @param ruleSet 规则集对象
* @param model 模型对象
* @param provider 原始provider对象
* @returns 解析出的provider对象
*/
export function provider2Provider<M extends MinimalModel, R extends MinimalProvider, P extends R = R>(
ruleSet: RuleSet<M, R>,
model: M,
provider: P
): P {
for (const rule of ruleSet.rules) {
if (rule.match(model)) {
return rule.provider(provider) as P
}
}
return ruleSet.fallbackRule(provider) as P
}

View File

@@ -0,0 +1,6 @@
export { aihubmixProviderCreator } from './aihubmix'
export { azureAnthropicProviderCreator } from './azure-anthropic'
export { endpointIs, provider2Provider, startsWith } from './helper'
export { newApiResolverCreator } from './newApi'
export type { RuleSet } from './types'
export { vertexAnthropicProviderCreator } from './vertex-anthropic'

View File

@@ -1,8 +1,7 @@
/**
* NewAPI规则集
*/
import type { Provider } from '@renderer/types'
import type { MinimalModel, MinimalProvider, ProviderType } from '../types'
import { endpointIs, provider2Provider } from './helper'
import type { RuleSet } from './types'
@@ -10,42 +9,43 @@ const NEWAPI_RULES: RuleSet = {
rules: [
{
match: endpointIs('anthropic'),
provider: (provider: Provider) => {
provider: (provider) => {
return {
...provider,
type: 'anthropic'
type: 'anthropic' as ProviderType
}
}
},
{
match: endpointIs('gemini'),
provider: (provider: Provider) => {
provider: (provider) => {
return {
...provider,
type: 'gemini'
type: 'gemini' as ProviderType
}
}
},
{
match: endpointIs('openai-response'),
provider: (provider: Provider) => {
provider: (provider) => {
return {
...provider,
type: 'openai-response'
type: 'openai-response' as ProviderType
}
}
},
{
match: (model) => endpointIs('openai')(model) || endpointIs('image-generation')(model),
provider: (provider: Provider) => {
provider: (provider) => {
return {
...provider,
type: 'openai'
type: 'openai' as ProviderType
}
}
}
],
fallbackRule: (provider: Provider) => provider
fallbackRule: (provider) => provider
}
export const newApiResolverCreator = provider2Provider.bind(null, NEWAPI_RULES)
export const newApiResolverCreator = <P extends MinimalProvider>(model: MinimalModel, provider: P): P =>
provider2Provider<MinimalModel, MinimalProvider, P>(NEWAPI_RULES, model, provider)

View File

@@ -0,0 +1,9 @@
import type { MinimalModel, MinimalProvider } from '../types'
export interface RuleSet<M extends MinimalModel = MinimalModel, P extends MinimalProvider = MinimalProvider> {
rules: Array<{
match: (model: M) => boolean
provider: (provider: P) => P
}>
fallbackRule: (provider: P) => P
}

View File

@@ -0,0 +1,19 @@
import type { MinimalModel, MinimalProvider } from '../types'
import { provider2Provider, startsWith } from './helper'
import type { RuleSet } from './types'
const VERTEX_ANTHROPIC_RULES: RuleSet = {
rules: [
{
match: startsWith('claude'),
provider: (provider: MinimalProvider) => ({
...provider,
id: 'google-vertex-anthropic'
})
}
],
fallbackRule: (provider: MinimalProvider) => provider
}
export const vertexAnthropicProviderCreator = <P extends MinimalProvider>(model: MinimalModel, provider: P): P =>
provider2Provider<MinimalModel, MinimalProvider, P>(VERTEX_ANTHROPIC_RULES, model, provider)

View File

@@ -0,0 +1,26 @@
import { getLowerBaseModelName } from '@shared/utils/naming'
import type { MinimalModel } from './types'
export const COPILOT_EDITOR_VERSION = 'vscode/1.104.1'
export const COPILOT_PLUGIN_VERSION = 'copilot-chat/0.26.7'
export const COPILOT_INTEGRATION_ID = 'vscode-chat'
export const COPILOT_USER_AGENT = 'GitHubCopilotChat/0.26.7'
export const COPILOT_DEFAULT_HEADERS = {
'Copilot-Integration-Id': COPILOT_INTEGRATION_ID,
'User-Agent': COPILOT_USER_AGENT,
'Editor-Version': COPILOT_EDITOR_VERSION,
'Editor-Plugin-Version': COPILOT_PLUGIN_VERSION,
'editor-version': COPILOT_EDITOR_VERSION,
'editor-plugin-version': COPILOT_PLUGIN_VERSION,
'copilot-vision-request': 'true'
} as const
// Models that require the OpenAI Responses endpoint when routed through GitHub Copilot (#10560)
const COPILOT_RESPONSES_MODEL_IDS = ['gpt-5-codex', 'gpt-5.1-codex', 'gpt-5.1-codex-mini']
export function isCopilotResponsesModel<M extends MinimalModel>(model: M): boolean {
const normalizedId = getLowerBaseModelName(model.id)
return COPILOT_RESPONSES_MODEL_IDS.some((target) => normalizedId === target)
}

View File

@@ -0,0 +1,100 @@
/**
* Provider Type Detection Utilities
*
* Functions to detect provider types based on provider configuration.
* These are pure functions that only depend on provider.type and provider.id.
*
* NOTE: These functions should match the logic in @renderer/utils/provider.ts
*/
import type { MinimalProvider } from './types'
/**
* Check if provider is Anthropic type
*/
export function isAnthropicProvider<P extends MinimalProvider>(provider: P): boolean {
return provider.type === 'anthropic'
}
/**
* Check if provider is OpenAI Response type (openai-response)
* NOTE: This matches isOpenAIProvider in renderer/utils/provider.ts
*/
export function isOpenAIProvider<P extends MinimalProvider>(provider: P): boolean {
return provider.type === 'openai-response'
}
/**
* Check if provider is Gemini type
*/
export function isGeminiProvider<P extends MinimalProvider>(provider: P): boolean {
return provider.type === 'gemini'
}
/**
* Check if provider is Azure OpenAI type
*/
export function isAzureOpenAIProvider<P extends MinimalProvider>(provider: P): boolean {
return provider.type === 'azure-openai'
}
/**
* Check if provider is Vertex AI type
*/
export function isVertexProvider<P extends MinimalProvider>(provider: P): boolean {
return provider.type === 'vertexai'
}
/**
* Check if provider is AWS Bedrock type
*/
export function isAwsBedrockProvider<P extends MinimalProvider>(provider: P): boolean {
return provider.type === 'aws-bedrock'
}
/**
* Check if provider is AI Gateway type
*/
export function isAIGatewayProvider<P extends MinimalProvider>(provider: P): boolean {
return provider.type === 'ai-gateway'
}
/**
* Check if Azure OpenAI provider uses responses endpoint
* Matches isAzureResponsesEndpoint in renderer/utils/provider.ts
*/
export function isAzureResponsesEndpoint<P extends MinimalProvider>(provider: P): boolean {
return provider.apiVersion === 'preview' || provider.apiVersion === 'v1'
}
/**
* Check if provider is Cherry AI type
* Matches isCherryAIProvider in renderer/utils/provider.ts
*/
export function isCherryAIProvider<P extends MinimalProvider>(provider: P): boolean {
return provider.id === 'cherryai'
}
/**
* Check if provider is Perplexity type
* Matches isPerplexityProvider in renderer/utils/provider.ts
*/
export function isPerplexityProvider<P extends MinimalProvider>(provider: P): boolean {
return provider.id === 'perplexity'
}
/**
* Check if provider is new-api type (supports multiple backends)
* Matches isNewApiProvider in renderer/utils/provider.ts
*/
export function isNewApiProvider<P extends MinimalProvider>(provider: P): boolean {
return ['new-api', 'cherryin'].includes(provider.id) || provider.type === ('new-api' as string)
}
/**
* Check if provider is OpenAI compatible
* Matches isOpenAICompatibleProvider in renderer/utils/provider.ts
*/
export function isOpenAICompatibleProvider<P extends MinimalProvider>(provider: P): boolean {
return ['openai', 'new-api', 'mistral'].includes(provider.type)
}

View File

@@ -0,0 +1,136 @@
/**
* Provider API Host Formatting
*
* Utilities for formatting provider API hosts to work with AI SDK.
* These handle the differences between how Cherry Studio stores API hosts
* and how AI SDK expects them.
*/
import {
formatApiHost,
formatAzureOpenAIApiHost,
formatVertexApiHost,
routeToEndpoint,
withoutTrailingSlash
} from '../api'
import {
isAnthropicProvider,
isAzureOpenAIProvider,
isCherryAIProvider,
isGeminiProvider,
isPerplexityProvider,
isVertexProvider
} from './detection'
import type { MinimalProvider } from './types'
import { SystemProviderIds } from './types'
/**
* Interface for environment-specific implementations
* Renderer and Main process can provide their own implementations
*/
export interface ProviderFormatContext {
vertex: {
project: string
location: string
}
}
/**
* Default Azure OpenAI API host formatter
*/
export function defaultFormatAzureOpenAIApiHost(host: string): string {
const normalizedHost = withoutTrailingSlash(host)
?.replace(/\/v1$/, '')
.replace(/\/openai$/, '')
// AI SDK will add /v1
return formatApiHost(normalizedHost + '/openai', false)
}
/**
* Format provider API host for AI SDK
*
* This function normalizes the apiHost to work with AI SDK.
* Different providers have different requirements:
* - Most providers: add /v1 suffix
* - Gemini: add /v1beta suffix
* - Some providers: no suffix needed
*
* @param provider - The provider to format
* @param context - Optional context with environment-specific implementations
* @returns Provider with formatted apiHost (and anthropicApiHost if applicable)
*/
export function formatProviderApiHost<T extends MinimalProvider>(provider: T, context: ProviderFormatContext): T {
const formatted = { ...provider }
// Format anthropicApiHost if present
if (formatted.anthropicApiHost) {
formatted.anthropicApiHost = formatApiHost(formatted.anthropicApiHost)
}
// Format based on provider type
if (isAnthropicProvider(provider)) {
const baseHost = formatted.anthropicApiHost || formatted.apiHost
// AI SDK needs /v1 in baseURL
formatted.apiHost = formatApiHost(baseHost)
if (!formatted.anthropicApiHost) {
formatted.anthropicApiHost = formatted.apiHost
}
} else if (formatted.id === SystemProviderIds.copilot || formatted.id === SystemProviderIds.github) {
formatted.apiHost = formatApiHost(formatted.apiHost, false)
} else if (isGeminiProvider(formatted)) {
formatted.apiHost = formatApiHost(formatted.apiHost, true, 'v1beta')
} else if (isAzureOpenAIProvider(formatted)) {
formatted.apiHost = formatAzureOpenAIApiHost(formatted.apiHost)
} else if (isVertexProvider(formatted)) {
formatted.apiHost = formatVertexApiHost(formatted, context.vertex.project, context.vertex.location)
} else if (isCherryAIProvider(formatted)) {
formatted.apiHost = formatApiHost(formatted.apiHost, false)
} else if (isPerplexityProvider(formatted)) {
formatted.apiHost = formatApiHost(formatted.apiHost, false)
} else {
formatted.apiHost = formatApiHost(formatted.apiHost)
}
return formatted
}
/**
* Get the base URL for AI SDK from a formatted provider
*
* This extracts the baseURL that AI SDK expects, handling
* the '#' endpoint routing format if present.
*
* @param formattedApiHost - The formatted apiHost (after formatProviderApiHost)
* @returns The baseURL for AI SDK
*/
export function getBaseUrlForAiSdk(formattedApiHost: string): string {
const { baseURL } = routeToEndpoint(formattedApiHost)
return baseURL
}
/**
* Get rotated API key from comma-separated keys
*
* This is the interface for API key rotation. The actual implementation
* depends on the environment (renderer uses window.keyv, main uses its own storage).
*/
export interface ApiKeyRotator {
/**
* Get the next API key in rotation
* @param providerId - The provider ID for tracking rotation
* @param keys - Comma-separated API keys
* @returns The next API key to use
*/
getRotatedKey(providerId: string, keys: string): string
}
/**
* Simple API key rotator that always returns the first key
* Use this when rotation is not needed
*/
export const simpleKeyRotator: ApiKeyRotator = {
getRotatedKey(_providerId: string, keys: string): string {
const keyList = keys.split(',').map((k) => k.trim())
return keyList[0] || keys
}
}

View File

@@ -0,0 +1,48 @@
/**
* Shared Provider Utilities
*
* This module exports utilities for working with AI providers
* that can be shared between main process and renderer process.
*/
// Type definitions
export type { MinimalProvider, ProviderType, SystemProviderId } from './types'
export { SystemProviderIds } from './types'
// Provider type detection
export {
isAIGatewayProvider,
isAnthropicProvider,
isAwsBedrockProvider,
isAzureOpenAIProvider,
isAzureResponsesEndpoint,
isCherryAIProvider,
isGeminiProvider,
isNewApiProvider,
isOpenAICompatibleProvider,
isOpenAIProvider,
isPerplexityProvider,
isVertexProvider
} from './detection'
// API host formatting
export type { ApiKeyRotator, ProviderFormatContext } from './format'
export {
defaultFormatAzureOpenAIApiHost,
formatProviderApiHost,
getBaseUrlForAiSdk,
simpleKeyRotator
} from './format'
// Provider ID mapping
export { getAiSdkProviderId, STATIC_PROVIDER_MAPPING, tryResolveProviderId } from './mapping'
// AI SDK configuration
export type { AiSdkConfig, AiSdkConfigContext } from './sdk-config'
export { providerToAiSdkConfig } from './sdk-config'
// Provider resolution
export { resolveActualProvider } from './resolve'
// Provider initialization
export { initializeSharedProviders, SHARED_PROVIDER_CONFIGS } from './initialization'

View File

@@ -0,0 +1,107 @@
import { type ProviderConfig, registerMultipleProviderConfigs } from '@cherrystudio/ai-core/provider'
type ProviderInitializationLogger = {
warn?: (message: string) => void
error?: (message: string, error: Error) => void
}
export const SHARED_PROVIDER_CONFIGS: ProviderConfig[] = [
{
id: 'openrouter',
name: 'OpenRouter',
import: () => import('@openrouter/ai-sdk-provider'),
creatorFunctionName: 'createOpenRouter',
supportsImageGeneration: true,
aliases: ['openrouter']
},
{
id: 'google-vertex',
name: 'Google Vertex AI',
import: () => import('@ai-sdk/google-vertex/edge'),
creatorFunctionName: 'createVertex',
supportsImageGeneration: true,
aliases: ['vertexai']
},
{
id: 'google-vertex-anthropic',
name: 'Google Vertex AI Anthropic',
import: () => import('@ai-sdk/google-vertex/anthropic/edge'),
creatorFunctionName: 'createVertexAnthropic',
supportsImageGeneration: true,
aliases: ['vertexai-anthropic']
},
{
id: 'azure-anthropic',
name: 'Azure AI Anthropic',
import: () => import('@ai-sdk/anthropic'),
creatorFunctionName: 'createAnthropic',
supportsImageGeneration: false,
aliases: ['azure-anthropic']
},
{
id: 'github-copilot-openai-compatible',
name: 'GitHub Copilot OpenAI Compatible',
import: () => import('@opeoginni/github-copilot-openai-compatible'),
creatorFunctionName: 'createGitHubCopilotOpenAICompatible',
supportsImageGeneration: false,
aliases: ['copilot', 'github-copilot']
},
{
id: 'bedrock',
name: 'Amazon Bedrock',
import: () => import('@ai-sdk/amazon-bedrock'),
creatorFunctionName: 'createAmazonBedrock',
supportsImageGeneration: true,
aliases: ['aws-bedrock']
},
{
id: 'perplexity',
name: 'Perplexity',
import: () => import('@ai-sdk/perplexity'),
creatorFunctionName: 'createPerplexity',
supportsImageGeneration: false,
aliases: ['perplexity']
},
{
id: 'mistral',
name: 'Mistral',
import: () => import('@ai-sdk/mistral'),
creatorFunctionName: 'createMistral',
supportsImageGeneration: false,
aliases: ['mistral']
},
{
id: 'huggingface',
name: 'HuggingFace',
import: () => import('@ai-sdk/huggingface'),
creatorFunctionName: 'createHuggingFace',
supportsImageGeneration: true,
aliases: ['hf', 'hugging-face']
},
{
id: 'ai-gateway',
name: 'AI Gateway',
import: () => import('@ai-sdk/gateway'),
creatorFunctionName: 'createGateway',
supportsImageGeneration: true,
aliases: ['gateway']
},
{
id: 'cerebras',
name: 'Cerebras',
import: () => import('@ai-sdk/cerebras'),
creatorFunctionName: 'createCerebras',
supportsImageGeneration: false
}
] as const
export function initializeSharedProviders(logger?: ProviderInitializationLogger): void {
try {
const successCount = registerMultipleProviderConfigs(SHARED_PROVIDER_CONFIGS)
if (successCount < SHARED_PROVIDER_CONFIGS.length) {
logger?.warn?.('Some providers failed to register. Check previous error logs.')
}
} catch (error) {
logger?.error?.('Failed to initialize shared providers', error as Error)
}
}

View File

@@ -0,0 +1,95 @@
/**
* Provider ID Mapping
*
* Maps Cherry Studio provider IDs/types to AI SDK provider IDs.
* This logic should match @renderer/aiCore/provider/factory.ts
*/
import { hasProviderConfigByAlias, type ProviderId, resolveProviderConfigId } from '@cherrystudio/ai-core/provider'
import { isAzureOpenAIProvider, isAzureResponsesEndpoint } from './detection'
import type { MinimalProvider } from './types'
/**
* Static mapping from Cherry Studio provider ID/type to AI SDK provider ID
* Matches STATIC_PROVIDER_MAPPING in @renderer/aiCore/provider/factory.ts
*/
export const STATIC_PROVIDER_MAPPING: Record<string, ProviderId> = {
gemini: 'google', // Google Gemini -> google
'azure-openai': 'azure', // Azure OpenAI -> azure
'openai-response': 'openai', // OpenAI Responses -> openai
grok: 'xai', // Grok -> xai
copilot: 'github-copilot-openai-compatible'
}
/**
* Try to resolve a provider identifier to an AI SDK provider ID
* Matches tryResolveProviderId in @renderer/aiCore/provider/factory.ts
*
* @param identifier - The provider ID or type to resolve
* @param checker - Provider config checker (defaults to static mapping only)
* @returns The resolved AI SDK provider ID, or null if not found
*/
export function tryResolveProviderId(identifier: string): ProviderId | null {
// 1. 检查静态映射
const staticMapping = STATIC_PROVIDER_MAPPING[identifier]
if (staticMapping) {
return staticMapping
}
// 2. 检查AiCore是否支持包括别名支持
if (hasProviderConfigByAlias(identifier)) {
// 解析为真实的Provider ID
return resolveProviderConfigId(identifier) as ProviderId
}
return null
}
/**
* Get the AI SDK Provider ID for a Cherry Studio provider
* Matches getAiSdkProviderId in @renderer/aiCore/provider/factory.ts
*
* Logic:
* 1. Handle Azure OpenAI specially (check responses endpoint)
* 2. Try to resolve from provider.id
* 3. Try to resolve from provider.type (but not for generic 'openai' type)
* 4. Check for OpenAI API host pattern
* 5. Fallback to provider's own ID
*
* @param provider - The Cherry Studio provider
* @param checker - Provider config checker (defaults to static mapping only)
* @returns The AI SDK provider ID to use
*/
export function getAiSdkProviderId(provider: MinimalProvider): ProviderId {
// 1. Handle Azure OpenAI specially - check this FIRST before other resolution
if (isAzureOpenAIProvider(provider)) {
if (isAzureResponsesEndpoint(provider)) {
return 'azure-responses'
}
return 'azure'
}
// 2. 尝试解析provider.id
const resolvedFromId = tryResolveProviderId(provider.id)
if (resolvedFromId) {
return resolvedFromId
}
// 3. 尝试解析provider.type
// 会把所有类型为openai的自定义provider解析到aisdk的openaiProvider上
if (provider.type !== 'openai') {
const resolvedFromType = tryResolveProviderId(provider.type)
if (resolvedFromType) {
return resolvedFromType
}
}
// 4. Check for OpenAI API host pattern
if (provider.apiHost.includes('api.openai.com')) {
return 'openai-chat'
}
// 5. 最后的fallback使用provider本身的id
return provider.id
}

View File

@@ -0,0 +1,43 @@
import { aihubmixProviderCreator, newApiResolverCreator, vertexAnthropicProviderCreator } from './config'
import { azureAnthropicProviderCreator } from './config/azure-anthropic'
import { isAzureOpenAIProvider, isNewApiProvider } from './detection'
import type { MinimalModel, MinimalProvider } from './types'
export interface ResolveActualProviderOptions<P extends MinimalProvider> {
isSystemProvider?: (provider: P) => boolean
}
const defaultIsSystemProvider = <P extends MinimalProvider>(provider: P): boolean => {
if ('isSystem' in provider) {
return Boolean((provider as unknown as { isSystem?: boolean }).isSystem)
}
return false
}
export function resolveActualProvider<M extends MinimalModel, P extends MinimalProvider>(
provider: P,
model: M,
options: ResolveActualProviderOptions<P> = {}
): P {
let resolvedProvider = provider
if (isNewApiProvider(resolvedProvider)) {
resolvedProvider = newApiResolverCreator(model, resolvedProvider)
}
const isSystemProvider = options.isSystemProvider?.(resolvedProvider) ?? defaultIsSystemProvider(resolvedProvider)
if (isSystemProvider && resolvedProvider.id === 'aihubmix') {
resolvedProvider = aihubmixProviderCreator(model, resolvedProvider)
}
if (isSystemProvider && resolvedProvider.id === 'vertexai') {
resolvedProvider = vertexAnthropicProviderCreator(model, resolvedProvider)
}
if (isAzureOpenAIProvider(resolvedProvider)) {
resolvedProvider = azureAnthropicProviderCreator(model, resolvedProvider)
}
return resolvedProvider
}

View File

@@ -0,0 +1,259 @@
/**
* AI SDK Configuration
*
* Shared utilities for converting Cherry Studio Provider to AI SDK configuration.
* Environment-specific logic (renderer/main) is injected via context interfaces.
*/
import { formatPrivateKey, hasProviderConfig, ProviderConfigFactory } from '@cherrystudio/ai-core/provider'
import { routeToEndpoint } from '../api'
import { getAiSdkProviderId } from './mapping'
import type { MinimalProvider } from './types'
import { SystemProviderIds } from './types'
/**
* AI SDK configuration result
*/
export interface AiSdkConfig {
providerId: string
options: Record<string, unknown>
}
/**
* Context for environment-specific implementations
*/
export interface AiSdkConfigContext {
/**
* Get the rotated API key (for multi-key support)
* Default: returns first key
*/
getRotatedApiKey?: (provider: MinimalProvider) => string
/**
* Check if a model uses chat completion only (for OpenAI response mode)
* Default: returns false
*/
isOpenAIChatCompletionOnlyModel?: (modelId: string) => boolean
/**
* Get Copilot default headers (constants)
* Default: returns empty object
*/
getCopilotDefaultHeaders?: () => Record<string, string>
/**
* Get Copilot stored headers from state
* Default: returns empty object
*/
getCopilotStoredHeaders?: () => Record<string, string>
/**
* Get AWS Bedrock configuration
* Default: returns undefined (not configured)
*/
getAwsBedrockConfig?: () =>
| {
authType: 'apiKey' | 'iam'
region: string
apiKey?: string
accessKeyId?: string
secretAccessKey?: string
}
| undefined
/**
* Get Vertex AI configuration
* Default: returns undefined (not configured)
*/
getVertexConfig?: (provider: MinimalProvider) =>
| {
project: string
location: string
googleCredentials: {
privateKey: string
clientEmail: string
}
}
| undefined
/**
* Get endpoint type for cherryin provider
*/
getEndpointType?: (modelId: string) => string | undefined
/**
* Custom fetch implementation
* Main process: use Electron net.fetch
* Renderer process: use browser fetch (default)
*/
fetch?: typeof globalThis.fetch
/**
* Get CherryAI signed fetch wrapper
* Returns a fetch function that adds signature headers to requests
*/
getCherryAISignedFetch?: () => typeof globalThis.fetch
}
/**
* Default simple key rotator - returns first key
*/
function defaultGetRotatedApiKey(provider: MinimalProvider): string {
const keys = provider.apiKey.split(',').map((k) => k.trim())
return keys[0] || provider.apiKey
}
/**
* Convert Cherry Studio Provider to AI SDK configuration
*
* @param provider - The formatted provider (after formatProviderApiHost)
* @param modelId - The model ID to use
* @param context - Environment-specific implementations
* @returns AI SDK configuration
*/
export function providerToAiSdkConfig(
provider: MinimalProvider,
modelId: string,
context: AiSdkConfigContext = {}
): AiSdkConfig {
const getRotatedApiKey = context.getRotatedApiKey || defaultGetRotatedApiKey
const isOpenAIChatCompletionOnlyModel = context.isOpenAIChatCompletionOnlyModel || (() => false)
const aiSdkProviderId = getAiSdkProviderId(provider)
// Build base config
const { baseURL, endpoint } = routeToEndpoint(provider.apiHost)
const baseConfig = {
baseURL,
apiKey: getRotatedApiKey(provider)
}
// Handle Copilot specially
if (provider.id === SystemProviderIds.copilot) {
const defaultHeaders = context.getCopilotDefaultHeaders?.() ?? {}
const storedHeaders = context.getCopilotStoredHeaders?.() ?? {}
const copilotExtraOptions: Record<string, unknown> = {
headers: {
...defaultHeaders,
...storedHeaders,
...provider.extra_headers
},
name: provider.id,
includeUsage: true
}
if (context.fetch) {
copilotExtraOptions.fetch = context.fetch
}
const options = ProviderConfigFactory.fromProvider(
'github-copilot-openai-compatible',
baseConfig,
copilotExtraOptions
)
return {
providerId: 'github-copilot-openai-compatible',
options
}
}
// Build extra options
const extraOptions: Record<string, unknown> = {}
if (endpoint) {
extraOptions.endpoint = endpoint
}
// Handle OpenAI mode
if (provider.type === 'openai-response' && !isOpenAIChatCompletionOnlyModel(modelId)) {
extraOptions.mode = 'responses'
} else if (aiSdkProviderId === 'openai' || (aiSdkProviderId === 'cherryin' && provider.type === 'openai')) {
extraOptions.mode = 'chat'
}
// Add extra headers
if (provider.extra_headers) {
extraOptions.headers = provider.extra_headers
if (aiSdkProviderId === 'openai') {
extraOptions.headers = {
...(extraOptions.headers as Record<string, string>),
'HTTP-Referer': 'https://cherry-ai.com',
'X-Title': 'Cherry Studio',
'X-Api-Key': baseConfig.apiKey
}
}
}
// Handle Azure modes
if (aiSdkProviderId === 'azure-responses') {
extraOptions.mode = 'responses'
} else if (aiSdkProviderId === 'azure') {
extraOptions.mode = 'chat'
}
// Handle AWS Bedrock
if (aiSdkProviderId === 'bedrock') {
const bedrockConfig = context.getAwsBedrockConfig?.()
if (bedrockConfig) {
extraOptions.region = bedrockConfig.region
if (bedrockConfig.authType === 'apiKey') {
extraOptions.apiKey = bedrockConfig.apiKey
} else {
extraOptions.accessKeyId = bedrockConfig.accessKeyId
extraOptions.secretAccessKey = bedrockConfig.secretAccessKey
}
}
}
// Handle Vertex AI
if (aiSdkProviderId === 'google-vertex' || aiSdkProviderId === 'google-vertex-anthropic') {
const vertexConfig = context.getVertexConfig?.(provider)
if (vertexConfig) {
extraOptions.project = vertexConfig.project
extraOptions.location = vertexConfig.location
extraOptions.googleCredentials = {
...vertexConfig.googleCredentials,
privateKey: formatPrivateKey(vertexConfig.googleCredentials.privateKey)
}
baseConfig.baseURL += aiSdkProviderId === 'google-vertex' ? '/publishers/google' : '/publishers/anthropic/models'
}
}
// Handle cherryin endpoint type
if (aiSdkProviderId === 'cherryin') {
const endpointType = context.getEndpointType?.(modelId)
if (endpointType) {
extraOptions.endpointType = endpointType
}
}
// Handle cherryai signed fetch
if (provider.id === 'cherryai') {
const signedFetch = context.getCherryAISignedFetch?.()
if (signedFetch) {
extraOptions.fetch = signedFetch
}
} else if (context.fetch) {
extraOptions.fetch = context.fetch
}
// Check if AI SDK supports this provider natively
if (hasProviderConfig(aiSdkProviderId) && aiSdkProviderId !== 'openai-compatible') {
const options = ProviderConfigFactory.fromProvider(aiSdkProviderId, baseConfig, extraOptions)
return {
providerId: aiSdkProviderId,
options
}
}
// Fallback to openai-compatible
const options = ProviderConfigFactory.createOpenAICompatible(baseConfig.baseURL, baseConfig.apiKey)
return {
providerId: 'openai-compatible',
options: {
...options,
name: provider.id,
...extraOptions,
includeUsage: true
}
}
}

View File

@@ -0,0 +1,174 @@
import * as z from 'zod'
export const ProviderTypeSchema = z.enum([
'openai',
'openai-response',
'anthropic',
'gemini',
'azure-openai',
'vertexai',
'mistral',
'aws-bedrock',
'vertex-anthropic',
'new-api',
'ai-gateway'
])
export type ProviderType = z.infer<typeof ProviderTypeSchema>
/**
* Minimal provider interface for shared utilities
* This is the subset of Provider that shared code needs
*/
export type MinimalProvider = {
id: string
type: ProviderType
apiKey: string
apiHost: string
anthropicApiHost?: string
apiVersion?: string
extra_headers?: Record<string, string>
}
/**
* Minimal model interface for shared utilities
* This is the subset of Model that shared code needs
*/
export type MinimalModel = {
id: string
endpoint_type?: string
}
export const SystemProviderIdSchema = z.enum([
'cherryin',
'silicon',
'aihubmix',
'ocoolai',
'deepseek',
'ppio',
'alayanew',
'qiniu',
'dmxapi',
'burncloud',
'tokenflux',
'302ai',
'cephalon',
'lanyun',
'ph8',
'openrouter',
'ollama',
'ovms',
'new-api',
'lmstudio',
'anthropic',
'openai',
'azure-openai',
'gemini',
'vertexai',
'github',
'copilot',
'zhipu',
'yi',
'moonshot',
'baichuan',
'dashscope',
'stepfun',
'doubao',
'infini',
'minimax',
'groq',
'together',
'fireworks',
'nvidia',
'grok',
'hyperbolic',
'mistral',
'jina',
'perplexity',
'modelscope',
'xirang',
'hunyuan',
'tencent-cloud-ti',
'baidu-cloud',
'gpustack',
'voyageai',
'aws-bedrock',
'poe',
'aionly',
'longcat',
'huggingface',
'sophnet',
'ai-gateway',
'cerebras'
])
export type SystemProviderId = z.infer<typeof SystemProviderIdSchema>
export const isSystemProviderId = (id: string): id is SystemProviderId => {
return SystemProviderIdSchema.safeParse(id).success
}
export const SystemProviderIds = {
cherryin: 'cherryin',
silicon: 'silicon',
aihubmix: 'aihubmix',
ocoolai: 'ocoolai',
deepseek: 'deepseek',
ppio: 'ppio',
alayanew: 'alayanew',
qiniu: 'qiniu',
dmxapi: 'dmxapi',
burncloud: 'burncloud',
tokenflux: 'tokenflux',
'302ai': '302ai',
cephalon: 'cephalon',
lanyun: 'lanyun',
ph8: 'ph8',
sophnet: 'sophnet',
openrouter: 'openrouter',
ollama: 'ollama',
ovms: 'ovms',
'new-api': 'new-api',
lmstudio: 'lmstudio',
anthropic: 'anthropic',
openai: 'openai',
'azure-openai': 'azure-openai',
gemini: 'gemini',
vertexai: 'vertexai',
github: 'github',
copilot: 'copilot',
zhipu: 'zhipu',
yi: 'yi',
moonshot: 'moonshot',
baichuan: 'baichuan',
dashscope: 'dashscope',
stepfun: 'stepfun',
doubao: 'doubao',
infini: 'infini',
minimax: 'minimax',
groq: 'groq',
together: 'together',
fireworks: 'fireworks',
nvidia: 'nvidia',
grok: 'grok',
hyperbolic: 'hyperbolic',
mistral: 'mistral',
jina: 'jina',
perplexity: 'perplexity',
modelscope: 'modelscope',
xirang: 'xirang',
hunyuan: 'hunyuan',
'tencent-cloud-ti': 'tencent-cloud-ti',
'baidu-cloud': 'baidu-cloud',
gpustack: 'gpustack',
voyageai: 'voyageai',
'aws-bedrock': 'aws-bedrock',
poe: 'poe',
aionly: 'aionly',
longcat: 'longcat',
huggingface: 'huggingface',
'ai-gateway': 'ai-gateway',
cerebras: 'cerebras'
} as const satisfies Record<SystemProviderId, SystemProviderId>
export type SystemProviderIdTypeMap = typeof SystemProviderIds

View File

@@ -0,0 +1 @@
export { getBaseModelName, getLowerBaseModelName } from './naming'

View File

@@ -0,0 +1,31 @@
/**
* 从模型 ID 中提取基础名称。
* 例如:
* - 'deepseek/deepseek-r1' => 'deepseek-r1'
* - 'deepseek-ai/deepseek/deepseek-r1' => 'deepseek-r1'
* @param {string} id 模型 ID
* @param {string} [delimiter='/'] 分隔符,默认为 '/'
* @returns {string} 基础名称
*/
export const getBaseModelName = (id: string, delimiter: string = '/'): string => {
const parts = id.split(delimiter)
return parts[parts.length - 1]
}
/**
* 从模型 ID 中提取基础名称并转换为小写。
* 例如:
* - 'deepseek/DeepSeek-R1' => 'deepseek-r1'
* - 'deepseek-ai/deepseek/DeepSeek-R1' => 'deepseek-r1'
* @param {string} id 模型 ID
* @param {string} [delimiter='/'] 分隔符,默认为 '/'
* @returns {string} 小写的基础名称
*/
export const getLowerBaseModelName = (id: string, delimiter: string = '/'): string => {
const baseModelName = getBaseModelName(id, delimiter).toLowerCase()
// for openrouter
if (baseModelName.endsWith(':free')) {
return baseModelName.replace(':free', '')
}
return baseModelName
}

View File

@@ -1,42 +1,64 @@
import { defineConfig, devices } from '@playwright/test'
import { defineConfig } from '@playwright/test'
/**
* See https://playwright.dev/docs/test-configuration.
* Playwright configuration for Electron e2e testing.
* See https://playwright.dev/docs/test-configuration
*/
export default defineConfig({
// Look for test files, relative to this configuration file.
testDir: './tests/e2e',
/* Run tests in files in parallel */
fullyParallel: true,
/* Fail the build on CI if you accidentally left test.only in the source code. */
forbidOnly: !!process.env.CI,
/* Retry on CI only */
retries: process.env.CI ? 2 : 0,
/* Opt out of parallel tests on CI. */
workers: process.env.CI ? 1 : undefined,
/* Reporter to use. See https://playwright.dev/docs/test-reporters */
reporter: 'html',
/* Shared settings for all the projects below. See https://playwright.dev/docs/api/class-testoptions. */
use: {
/* Base URL to use in actions like `await page.goto('/')`. */
// baseURL: 'http://localhost:3000',
// Look for test files in the specs directory
testDir: './tests/e2e/specs',
/* Collect trace when retrying the failed test. See https://playwright.dev/docs/trace-viewer */
trace: 'on-first-retry'
// Global timeout for each test
timeout: 60000,
// Assertion timeout
expect: {
timeout: 10000
},
/* Configure projects for major browsers */
// Electron apps should run tests sequentially to avoid conflicts
fullyParallel: false,
workers: 1,
// Fail the build on CI if you accidentally left test.only in the source code
forbidOnly: !!process.env.CI,
// Retry on CI only
retries: process.env.CI ? 2 : 0,
// Reporter configuration
reporter: [['html', { outputFolder: 'playwright-report' }], ['list']],
// Global setup and teardown
globalSetup: './tests/e2e/global-setup.ts',
globalTeardown: './tests/e2e/global-teardown.ts',
// Output directory for test artifacts
outputDir: './test-results',
// Shared settings for all tests
use: {
// Collect trace when retrying the failed test
trace: 'retain-on-failure',
// Take screenshot only on failure
screenshot: 'only-on-failure',
// Record video only on failure
video: 'retain-on-failure',
// Action timeout
actionTimeout: 15000,
// Navigation timeout
navigationTimeout: 30000
},
// Single project for Electron testing
projects: [
{
name: 'chromium',
use: { ...devices['Desktop Chrome'] }
name: 'electron',
testMatch: '**/*.spec.ts'
}
]
/* Run your local dev server before starting the tests */
// webServer: {
// command: 'npm run start',
// url: 'http://localhost:3000',
// reuseExistingServer: !process.env.CI,
// },
})

View File

@@ -0,0 +1,638 @@
/**
* AI SDK to Anthropic SSE Adapter
*
* Converts AI SDK's fullStream (TextStreamPart) events to Anthropic Messages API SSE format.
* This enables any AI provider supported by AI SDK to be exposed via Anthropic-compatible API.
*
* Anthropic SSE Event Flow:
* 1. message_start - Initial message with metadata
* 2. content_block_start - Begin a content block (text, tool_use, thinking)
* 3. content_block_delta - Incremental content updates
* 4. content_block_stop - End a content block
* 5. message_delta - Updates to overall message (stop_reason, usage)
* 6. message_stop - Stream complete
*
* @see https://docs.anthropic.com/en/api/messages-streaming
*/
import type {
ContentBlock,
InputJSONDelta,
Message,
MessageDeltaUsage,
RawContentBlockDeltaEvent,
RawContentBlockStartEvent,
RawContentBlockStopEvent,
RawMessageDeltaEvent,
RawMessageStartEvent,
RawMessageStopEvent,
RawMessageStreamEvent,
StopReason,
TextBlock,
TextDelta,
ThinkingBlock,
ThinkingDelta,
ToolUseBlock,
Usage
} from '@anthropic-ai/sdk/resources/messages'
import { loggerService } from '@logger'
import { type FinishReason, type LanguageModelUsage, type TextStreamPart, type ToolSet } from 'ai'
import { googleReasoningCache, openRouterReasoningCache } from '../../services/CacheService'
const logger = loggerService.withContext('AiSdkToAnthropicSSE')
interface ContentBlockState {
type: 'text' | 'tool_use' | 'thinking'
index: number
started: boolean
content: string
// For tool_use blocks
toolId?: string
toolName?: string
toolInput?: string
}
interface AdapterState {
messageId: string
model: string
inputTokens: number
outputTokens: number
cacheInputTokens: number
currentBlockIndex: number
blocks: Map<number, ContentBlockState>
textBlockIndex: number | null
// Track multiple thinking blocks by their reasoning ID
thinkingBlocks: Map<string, number> // reasoningId -> blockIndex
currentThinkingId: string | null // Currently active thinking block ID
toolBlocks: Map<string, number> // toolCallId -> blockIndex
stopReason: StopReason | null
hasEmittedMessageStart: boolean
}
export type SSEEventCallback = (event: RawMessageStreamEvent) => void
export interface AiSdkToAnthropicSSEOptions {
model: string
messageId?: string
inputTokens?: number
onEvent: SSEEventCallback
}
/**
* Adapter that converts AI SDK fullStream events to Anthropic SSE events
*/
export class AiSdkToAnthropicSSE {
private state: AdapterState
private onEvent: SSEEventCallback
constructor(options: AiSdkToAnthropicSSEOptions) {
this.onEvent = options.onEvent
this.state = {
messageId: options.messageId || `msg_${Date.now()}_${Math.random().toString(36).substring(2, 11)}`,
model: options.model,
inputTokens: options.inputTokens || 0,
outputTokens: 0,
cacheInputTokens: 0,
currentBlockIndex: 0,
blocks: new Map(),
textBlockIndex: null,
thinkingBlocks: new Map(),
currentThinkingId: null,
toolBlocks: new Map(),
stopReason: null,
hasEmittedMessageStart: false
}
}
/**
* Process the AI SDK stream and emit Anthropic SSE events
*/
async processStream(fullStream: ReadableStream<TextStreamPart<ToolSet>>): Promise<void> {
const reader = fullStream.getReader()
try {
// Emit message_start at the beginning
this.emitMessageStart()
while (true) {
const { done, value } = await reader.read()
if (done) {
break
}
this.processChunk(value)
}
// Ensure all blocks are closed and emit final events
this.finalize()
} catch (error) {
await reader.cancel()
throw error
} finally {
reader.releaseLock()
}
}
/**
* Process a single AI SDK chunk and emit corresponding Anthropic events
*/
private processChunk(chunk: TextStreamPart<ToolSet>): void {
logger.silly('AiSdkToAnthropicSSE - Processing chunk:', { chunk: JSON.stringify(chunk) })
switch (chunk.type) {
// === Text Events ===
case 'text-start':
this.startTextBlock()
break
case 'text-delta':
this.emitTextDelta(chunk.text || '')
break
case 'text-end':
this.stopTextBlock()
break
// === Reasoning/Thinking Events ===
case 'reasoning-start': {
const reasoningId = chunk.id
this.startThinkingBlock(reasoningId)
break
}
case 'reasoning-delta': {
const reasoningId = chunk.id
this.emitThinkingDelta(chunk.text || '', reasoningId)
break
}
case 'reasoning-end': {
const reasoningId = chunk.id
this.stopThinkingBlock(reasoningId)
break
}
// === Tool Events ===
case 'tool-call':
if (googleReasoningCache && chunk.providerMetadata?.google?.thoughtSignature) {
googleReasoningCache.set(
`google-${chunk.toolName}`,
chunk.providerMetadata?.google?.thoughtSignature as string
)
}
// FIXME: 按toolcall id绑定
if (
openRouterReasoningCache &&
chunk.providerMetadata?.openrouter?.reasoning_details &&
Array.isArray(chunk.providerMetadata.openrouter.reasoning_details)
) {
openRouterReasoningCache.set(
'openrouter',
JSON.parse(JSON.stringify(chunk.providerMetadata.openrouter.reasoning_details))
)
}
this.handleToolCall({
type: 'tool-call',
toolCallId: chunk.toolCallId,
toolName: chunk.toolName,
args: chunk.input
})
break
case 'tool-result':
// this.handleToolResult({
// type: 'tool-result',
// toolCallId: chunk.toolCallId,
// toolName: chunk.toolName,
// args: chunk.input,
// result: chunk.output
// })
break
case 'finish-step':
if (chunk.finishReason === 'tool-calls') {
this.state.stopReason = 'tool_use'
}
break
case 'finish':
this.handleFinish(chunk)
break
case 'error':
throw chunk.error
// Ignore other event types
default:
break
}
}
private emitMessageStart(): void {
if (this.state.hasEmittedMessageStart) return
this.state.hasEmittedMessageStart = true
const usage: Usage = {
input_tokens: this.state.inputTokens,
output_tokens: 0,
cache_creation_input_tokens: 0,
cache_read_input_tokens: 0,
server_tool_use: null
}
const message: Message = {
id: this.state.messageId,
type: 'message',
role: 'assistant',
content: [],
model: this.state.model,
stop_reason: null,
stop_sequence: null,
usage
}
const event: RawMessageStartEvent = {
type: 'message_start',
message
}
this.onEvent(event)
}
private startTextBlock(): void {
// If we already have a text block, don't create another
if (this.state.textBlockIndex !== null) return
const index = this.state.currentBlockIndex++
this.state.textBlockIndex = index
this.state.blocks.set(index, {
type: 'text',
index,
started: true,
content: ''
})
const contentBlock: TextBlock = {
type: 'text',
text: '',
citations: null
}
const event: RawContentBlockStartEvent = {
type: 'content_block_start',
index,
content_block: contentBlock
}
this.onEvent(event)
}
private emitTextDelta(text: string): void {
if (!text) return
// Auto-start text block if not started
if (this.state.textBlockIndex === null) {
this.startTextBlock()
}
const index = this.state.textBlockIndex!
const block = this.state.blocks.get(index)
if (block) {
block.content += text
}
const delta: TextDelta = {
type: 'text_delta',
text
}
const event: RawContentBlockDeltaEvent = {
type: 'content_block_delta',
index,
delta
}
this.onEvent(event)
}
private stopTextBlock(): void {
if (this.state.textBlockIndex === null) return
const index = this.state.textBlockIndex
const event: RawContentBlockStopEvent = {
type: 'content_block_stop',
index
}
this.onEvent(event)
this.state.textBlockIndex = null
}
private startThinkingBlock(reasoningId: string): void {
// Check if this thinking block already exists
if (this.state.thinkingBlocks.has(reasoningId)) return
const index = this.state.currentBlockIndex++
this.state.thinkingBlocks.set(reasoningId, index)
this.state.currentThinkingId = reasoningId
this.state.blocks.set(index, {
type: 'thinking',
index,
started: true,
content: ''
})
const contentBlock: ThinkingBlock = {
type: 'thinking',
thinking: '',
signature: ''
}
const event: RawContentBlockStartEvent = {
type: 'content_block_start',
index,
content_block: contentBlock
}
this.onEvent(event)
}
private emitThinkingDelta(text: string, reasoningId?: string): void {
if (!text) return
// Determine which thinking block to use
const targetId = reasoningId || this.state.currentThinkingId
if (!targetId) {
// Auto-start thinking block if not started
const newId = `reasoning_${Date.now()}`
this.startThinkingBlock(newId)
return this.emitThinkingDelta(text, newId)
}
const index = this.state.thinkingBlocks.get(targetId)
if (index === undefined) {
// If the block doesn't exist, create it
this.startThinkingBlock(targetId)
return this.emitThinkingDelta(text, targetId)
}
const block = this.state.blocks.get(index)
if (block) {
block.content += text
}
const delta: ThinkingDelta = {
type: 'thinking_delta',
thinking: text
}
const event: RawContentBlockDeltaEvent = {
type: 'content_block_delta',
index,
delta
}
this.onEvent(event)
}
private stopThinkingBlock(reasoningId?: string): void {
const targetId = reasoningId || this.state.currentThinkingId
if (!targetId) return
const index = this.state.thinkingBlocks.get(targetId)
if (index === undefined) return
const event: RawContentBlockStopEvent = {
type: 'content_block_stop',
index
}
this.onEvent(event)
this.state.thinkingBlocks.delete(targetId)
// Update currentThinkingId if we just closed the current one
if (this.state.currentThinkingId === targetId) {
// Set to the most recent remaining thinking block, or null if none
const remaining = Array.from(this.state.thinkingBlocks.keys())
this.state.currentThinkingId = remaining.length > 0 ? remaining[remaining.length - 1] : null
}
}
private handleToolCall(chunk: { type: 'tool-call'; toolCallId: string; toolName: string; args: unknown }): void {
const { toolCallId, toolName, args } = chunk
// Check if we already have this tool call
if (this.state.toolBlocks.has(toolCallId)) {
return
}
const index = this.state.currentBlockIndex++
this.state.toolBlocks.set(toolCallId, index)
const inputJson = JSON.stringify(args)
this.state.blocks.set(index, {
type: 'tool_use',
index,
started: true,
content: inputJson,
toolId: toolCallId,
toolName,
toolInput: inputJson
})
// Emit content_block_start for tool_use
const contentBlock: ToolUseBlock = {
type: 'tool_use',
id: toolCallId,
name: toolName,
input: {}
}
const startEvent: RawContentBlockStartEvent = {
type: 'content_block_start',
index,
content_block: contentBlock
}
this.onEvent(startEvent)
// Emit the full input as a delta (Anthropic streams JSON incrementally)
const delta: InputJSONDelta = {
type: 'input_json_delta',
partial_json: inputJson
}
const deltaEvent: RawContentBlockDeltaEvent = {
type: 'content_block_delta',
index,
delta
}
this.onEvent(deltaEvent)
// Emit content_block_stop
const stopEvent: RawContentBlockStopEvent = {
type: 'content_block_stop',
index
}
this.onEvent(stopEvent)
// Mark that we have tool use
this.state.stopReason = 'tool_use'
}
private handleFinish(chunk: { type: 'finish'; finishReason?: FinishReason; totalUsage?: LanguageModelUsage }): void {
// Update usage
if (chunk.totalUsage) {
this.state.inputTokens = chunk.totalUsage.inputTokens || 0
this.state.outputTokens = chunk.totalUsage.outputTokens || 0
this.state.cacheInputTokens = chunk.totalUsage.cachedInputTokens || 0
}
// Determine finish reason
if (!this.state.stopReason) {
switch (chunk.finishReason) {
case 'stop':
this.state.stopReason = 'end_turn'
break
case 'length':
this.state.stopReason = 'max_tokens'
break
case 'tool-calls':
this.state.stopReason = 'tool_use'
break
case 'content-filter':
this.state.stopReason = 'refusal'
break
default:
this.state.stopReason = 'end_turn'
}
}
}
private finalize(): void {
// Close any open blocks
if (this.state.textBlockIndex !== null) {
this.stopTextBlock()
}
// Close all open thinking blocks
for (const reasoningId of this.state.thinkingBlocks.keys()) {
this.stopThinkingBlock(reasoningId)
}
// Emit message_delta with final stop reason and usage
const usage: MessageDeltaUsage = {
output_tokens: this.state.outputTokens,
input_tokens: this.state.inputTokens,
cache_creation_input_tokens: this.state.cacheInputTokens,
cache_read_input_tokens: null,
server_tool_use: null
}
const messageDeltaEvent: RawMessageDeltaEvent = {
type: 'message_delta',
delta: {
stop_reason: this.state.stopReason || 'end_turn',
stop_sequence: null
},
usage
}
this.onEvent(messageDeltaEvent)
// Emit message_stop
const messageStopEvent: RawMessageStopEvent = {
type: 'message_stop'
}
this.onEvent(messageStopEvent)
}
/**
* Set input token count (typically from prompt)
*/
setInputTokens(count: number): void {
this.state.inputTokens = count
}
/**
* Get the current message ID
*/
getMessageId(): string {
return this.state.messageId
}
/**
* Build a complete Message object for non-streaming responses
*/
buildNonStreamingResponse(): Message {
const content: ContentBlock[] = []
// Collect all content blocks in order
const sortedBlocks = Array.from(this.state.blocks.values()).sort((a, b) => a.index - b.index)
for (const block of sortedBlocks) {
switch (block.type) {
case 'text':
content.push({
type: 'text',
text: block.content,
citations: null
} as TextBlock)
break
case 'thinking':
content.push({
type: 'thinking',
thinking: block.content
} as ThinkingBlock)
break
case 'tool_use':
content.push({
type: 'tool_use',
id: block.toolId!,
name: block.toolName!,
input: JSON.parse(block.toolInput || '{}')
} as ToolUseBlock)
break
}
}
return {
id: this.state.messageId,
type: 'message',
role: 'assistant',
content,
model: this.state.model,
stop_reason: this.state.stopReason || 'end_turn',
stop_sequence: null,
usage: {
input_tokens: this.state.inputTokens,
output_tokens: this.state.outputTokens,
cache_creation_input_tokens: 0,
cache_read_input_tokens: 0,
server_tool_use: null
}
}
}
}
/**
* Format an Anthropic SSE event for HTTP streaming
*/
export function formatSSEEvent(event: RawMessageStreamEvent): string {
return `event: ${event.type}\ndata: ${JSON.stringify(event)}\n\n`
}
/**
* Create a done marker for SSE stream
*/
export function formatSSEDone(): string {
return 'data: [DONE]\n\n'
}
export default AiSdkToAnthropicSSE

View File

@@ -0,0 +1,13 @@
/**
* Shared Adapters
*
* This module exports adapters for converting between different AI API formats.
*/
export {
AiSdkToAnthropicSSE,
type AiSdkToAnthropicSSEOptions,
formatSSEDone,
formatSSEEvent,
type SSEEventCallback
} from './AiSdkToAnthropicSSE'

View File

@@ -0,0 +1,95 @@
import * as z from 'zod/v4'
enum ReasoningFormat {
Unknown = 'unknown',
OpenAIResponsesV1 = 'openai-responses-v1',
XAIResponsesV1 = 'xai-responses-v1',
AnthropicClaudeV1 = 'anthropic-claude-v1',
GoogleGeminiV1 = 'google-gemini-v1'
}
// Anthropic Claude was the first reasoning that we're
// passing back and forth
export const DEFAULT_REASONING_FORMAT = ReasoningFormat.AnthropicClaudeV1
function isDefinedOrNotNull<T>(value: T | null | undefined): value is T {
return value !== null && value !== undefined
}
export enum ReasoningDetailType {
Summary = 'reasoning.summary',
Encrypted = 'reasoning.encrypted',
Text = 'reasoning.text'
}
export const CommonReasoningDetailSchema = z
.object({
id: z.string().nullish(),
format: z.enum(ReasoningFormat).nullish(),
index: z.number().optional()
})
.loose()
export const ReasoningDetailSummarySchema = z
.object({
type: z.literal(ReasoningDetailType.Summary),
summary: z.string()
})
.extend(CommonReasoningDetailSchema.shape)
export type ReasoningDetailSummary = z.infer<typeof ReasoningDetailSummarySchema>
export const ReasoningDetailEncryptedSchema = z
.object({
type: z.literal(ReasoningDetailType.Encrypted),
data: z.string()
})
.extend(CommonReasoningDetailSchema.shape)
export type ReasoningDetailEncrypted = z.infer<typeof ReasoningDetailEncryptedSchema>
export const ReasoningDetailTextSchema = z
.object({
type: z.literal(ReasoningDetailType.Text),
text: z.string().nullish(),
signature: z.string().nullish()
})
.extend(CommonReasoningDetailSchema.shape)
export type ReasoningDetailText = z.infer<typeof ReasoningDetailTextSchema>
export const ReasoningDetailUnionSchema = z.union([
ReasoningDetailSummarySchema,
ReasoningDetailEncryptedSchema,
ReasoningDetailTextSchema
])
export type ReasoningDetailUnion = z.infer<typeof ReasoningDetailUnionSchema>
const ReasoningDetailsWithUnknownSchema = z.union([ReasoningDetailUnionSchema, z.unknown().transform(() => null)])
export const ReasoningDetailArraySchema = z
.array(ReasoningDetailsWithUnknownSchema)
.transform((d) => d.filter((d): d is ReasoningDetailUnion => !!d))
export const OutputUnionToReasoningDetailsSchema = z.union([
z
.object({
delta: z.object({
reasoning_details: z.array(ReasoningDetailsWithUnknownSchema)
})
})
.transform((data) => data.delta.reasoning_details.filter(isDefinedOrNotNull)),
z
.object({
message: z.object({
reasoning_details: z.array(ReasoningDetailsWithUnknownSchema)
})
})
.transform((data) => data.message.reasoning_details.filter(isDefinedOrNotNull)),
z
.object({
text: z.string(),
reasoning_details: z.array(ReasoningDetailsWithUnknownSchema)
})
.transform((data) => data.reasoning_details.filter(isDefinedOrNotNull))
])

View File

@@ -1,17 +1,93 @@
import type { MessageCreateParams } from '@anthropic-ai/sdk/resources'
import { loggerService } from '@logger'
import { buildSharedMiddlewares, type SharedMiddlewareConfig } from '@shared/middleware'
import { getAiSdkProviderId } from '@shared/provider'
import type { Provider } from '@types'
import type { Request, Response } from 'express'
import express from 'express'
import { messagesService } from '../services/messages'
import { getProviderById, validateModelId } from '../utils'
import { generateUnifiedMessage, streamUnifiedMessages } from '../services/unified-messages'
import { getProviderById, isModelAnthropicCompatible, validateModelId } from '../utils'
/**
* Check if a specific model on a provider should use direct Anthropic SDK
*
* A provider+model combination is considered "Anthropic-compatible" if:
* 1. It's a native Anthropic provider (type === 'anthropic'), OR
* 2. It has anthropicApiHost configured AND the specific model supports Anthropic API
* (for aggregated providers like Silicon, only certain models support Anthropic endpoint)
*
* @param provider - The provider to check
* @param modelId - The model ID to check (without provider prefix)
* @returns true if should use direct Anthropic SDK, false for unified SDK
*/
function shouldUseDirectAnthropic(provider: Provider, modelId: string): boolean {
// Native Anthropic provider - always use direct SDK
if (provider.type === 'anthropic') {
return true
}
// No anthropicApiHost configured - use unified SDK
if (!provider.anthropicApiHost?.trim()) {
return false
}
// Has anthropicApiHost - check model-level compatibility
// For aggregated providers, only specific models support Anthropic API
return isModelAnthropicCompatible(provider, modelId)
}
const logger = loggerService.withContext('ApiServerMessagesRoutes')
const router = express.Router()
const providerRouter = express.Router({ mergeParams: true })
/**
* Estimate token count from messages
* Simple approximation: ~4 characters per token for English text
*/
interface CountTokensInput {
messages: Array<{ role: string; content: string | Array<{ type: string; text?: string }> }>
system?: string | Array<{ type: string; text?: string }>
}
function estimateTokenCount(input: CountTokensInput): number {
const { messages, system } = input
let totalChars = 0
// Count system message tokens
if (system) {
if (typeof system === 'string') {
totalChars += system.length
} else if (Array.isArray(system)) {
for (const block of system) {
if (block.type === 'text' && block.text) {
totalChars += block.text.length
}
}
}
}
// Count message tokens
for (const msg of messages) {
if (typeof msg.content === 'string') {
totalChars += msg.content.length
} else if (Array.isArray(msg.content)) {
for (const block of msg.content) {
if (block.type === 'text' && block.text) {
totalChars += block.text.length
}
}
}
// Add overhead for role
totalChars += 10
}
// Estimate tokens (~4 chars per token, with some overhead)
return Math.ceil(totalChars / 4) + messages.length * 3
}
// Helper function for basic request validation
async function validateRequestBody(req: Request): Promise<{ valid: boolean; error?: any }> {
const request: MessageCreateParams = req.body
@@ -33,21 +109,36 @@ async function validateRequestBody(req: Request): Promise<{ valid: boolean; erro
}
interface HandleMessageProcessingOptions {
req: Request
res: Response
provider: Provider
request: MessageCreateParams
modelId?: string
}
async function handleMessageProcessing({
req,
/**
* Handle message processing using direct Anthropic SDK
* Used for providers with anthropicApiHost or native Anthropic providers
* This bypasses AI SDK conversion and uses native Anthropic protocol
*/
async function handleDirectAnthropicProcessing({
res,
provider,
request,
modelId
}: HandleMessageProcessingOptions): Promise<void> {
modelId,
extraHeaders
}: HandleMessageProcessingOptions & { extraHeaders?: Record<string, string | string[]> }): Promise<void> {
const actualModelId = modelId || request.model
logger.info('Processing message via direct Anthropic SDK', {
providerId: provider.id,
providerType: provider.type,
modelId: actualModelId,
stream: !!request.stream,
anthropicApiHost: provider.anthropicApiHost
})
try {
// Validate request
const validation = messagesService.validateRequest(request)
if (!validation.isValid) {
res.status(400).json({
@@ -60,28 +151,126 @@ async function handleMessageProcessing({
return
}
const extraHeaders = messagesService.prepareHeaders(req.headers)
// Process message using messagesService (native Anthropic SDK)
const { client, anthropicRequest } = await messagesService.processMessage({
provider,
request,
extraHeaders,
modelId
modelId: actualModelId
})
if (request.stream) {
// Use native Anthropic streaming
await messagesService.handleStreaming(client, anthropicRequest, { response: res }, provider)
return
} else {
// Use native Anthropic non-streaming
const response = await client.messages.create(anthropicRequest)
res.json(response)
}
const response = await client.messages.create(anthropicRequest)
res.json(response)
} catch (error: any) {
logger.error('Message processing error', { error })
logger.error('Direct Anthropic processing error', { error })
const { statusCode, errorResponse } = messagesService.transformError(error)
res.status(statusCode).json(errorResponse)
}
}
/**
* Handle message processing using unified AI SDK
* Used for non-Anthropic providers that need format conversion
* - Uses AI SDK adapters with output converted to Anthropic SSE format
*/
async function handleUnifiedProcessing({
res,
provider,
request,
modelId
}: HandleMessageProcessingOptions): Promise<void> {
const actualModelId = modelId || request.model
logger.info('Processing message via unified AI SDK', {
providerId: provider.id,
providerType: provider.type,
modelId: actualModelId,
stream: !!request.stream
})
try {
// Validate request
const validation = messagesService.validateRequest(request)
if (!validation.isValid) {
res.status(400).json({
type: 'error',
error: {
type: 'invalid_request_error',
message: validation.errors.join('; ')
}
})
return
}
const middlewareConfig: SharedMiddlewareConfig = {
modelId: actualModelId,
providerId: provider.id,
aiSdkProviderId: getAiSdkProviderId(provider)
}
const middlewares = buildSharedMiddlewares(middlewareConfig)
logger.debug('Built middlewares for unified processing', {
middlewareCount: middlewares.length,
modelId: actualModelId,
providerId: provider.id
})
if (request.stream) {
await streamUnifiedMessages({
response: res,
provider,
modelId: actualModelId,
params: request,
middlewares,
onError: (error) => {
logger.error('Stream error', error as Error)
},
onComplete: () => {
logger.debug('Stream completed')
}
})
} else {
const response = await generateUnifiedMessage({
provider,
modelId: actualModelId,
params: request,
middlewares
})
res.json(response)
}
} catch (error: any) {
const { statusCode, errorResponse } = messagesService.transformError(error)
res.status(statusCode).json(errorResponse)
}
}
/**
* Handle message processing - routes to appropriate handler based on provider and model
*
* Routing logic:
* - Native Anthropic providers (type === 'anthropic'): Direct Anthropic SDK
* - Providers with anthropicApiHost AND model supports Anthropic API: Direct Anthropic SDK
* - Other providers/models: Unified AI SDK with Anthropic SSE conversion
*/
async function handleMessageProcessing({
res,
provider,
request,
modelId
}: HandleMessageProcessingOptions): Promise<void> {
const actualModelId = modelId || request.model
if (shouldUseDirectAnthropic(provider, actualModelId)) {
return handleDirectAnthropicProcessing({ res, provider, request, modelId })
}
return handleUnifiedProcessing({ res, provider, request, modelId })
}
/**
* @swagger
* /v1/messages:
@@ -235,7 +424,7 @@ router.post('/', async (req: Request, res: Response) => {
const provider = modelValidation.provider!
const modelId = modelValidation.modelId!
return handleMessageProcessing({ req, res, provider, request, modelId })
return handleMessageProcessing({ res, provider, request, modelId })
} catch (error: any) {
logger.error('Message processing error', { error })
const { statusCode, errorResponse } = messagesService.transformError(error)
@@ -393,7 +582,7 @@ providerRouter.post('/', async (req: Request, res: Response) => {
const request: MessageCreateParams = req.body
return handleMessageProcessing({ req, res, provider, request })
return handleMessageProcessing({ res, provider, request })
} catch (error: any) {
logger.error('Message processing error', { error })
const { statusCode, errorResponse } = messagesService.transformError(error)
@@ -401,4 +590,132 @@ providerRouter.post('/', async (req: Request, res: Response) => {
}
})
/**
* @swagger
* /v1/messages/count_tokens:
* post:
* summary: Count tokens for messages
* description: Count tokens for Anthropic Messages API format (required by Claude Code SDK)
* tags: [Messages]
* requestBody:
* required: true
* content:
* application/json:
* schema:
* type: object
* required:
* - model
* - messages
* properties:
* model:
* type: string
* description: Model ID
* messages:
* type: array
* items:
* type: object
* system:
* type: string
* description: System message
* responses:
* 200:
* description: Token count response
* content:
* application/json:
* schema:
* type: object
* properties:
* input_tokens:
* type: integer
* 400:
* description: Bad request
*/
router.post('/count_tokens', async (req: Request, res: Response) => {
try {
const { model, messages, system } = req.body
if (!model) {
return res.status(400).json({
type: 'error',
error: {
type: 'invalid_request_error',
message: 'model parameter is required'
}
})
}
if (!messages || !Array.isArray(messages)) {
return res.status(400).json({
type: 'error',
error: {
type: 'invalid_request_error',
message: 'messages parameter is required'
}
})
}
const estimatedTokens = estimateTokenCount({ messages, system })
logger.debug('Token count estimated', {
model,
messageCount: messages.length,
estimatedTokens
})
return res.json({
input_tokens: estimatedTokens
})
} catch (error: any) {
logger.error('Token counting error', { error })
return res.status(500).json({
type: 'error',
error: {
type: 'api_error',
message: error.message || 'Internal server error'
}
})
}
})
/**
* Provider-specific count_tokens endpoint
*/
providerRouter.post('/count_tokens', async (req: Request, res: Response) => {
try {
const { model, messages, system } = req.body
if (!messages || !Array.isArray(messages)) {
return res.status(400).json({
type: 'error',
error: {
type: 'invalid_request_error',
message: 'messages parameter is required'
}
})
}
const estimatedTokens = estimateTokenCount({ messages, system })
logger.debug('Token count estimated (provider route)', {
providerId: req.params.provider,
model,
messageCount: messages.length,
estimatedTokens
})
return res.json({
input_tokens: estimatedTokens
})
} catch (error: any) {
logger.error('Token counting error', { error })
return res.status(500).json({
type: 'error',
error: {
type: 'api_error',
message: error.message || 'Internal server error'
}
})
}
})
export { providerRouter as messagesProviderRoutes, router as messagesRoutes }

View File

@@ -2,8 +2,10 @@ import type Anthropic from '@anthropic-ai/sdk'
import type { MessageCreateParams, MessageStreamEvent } from '@anthropic-ai/sdk/resources'
import { loggerService } from '@logger'
import anthropicService from '@main/services/AnthropicService'
import { buildClaudeCodeSystemMessage, getSdkClient } from '@shared/anthropic'
import { buildClaudeCodeSystemMessage, getSdkClient, sanitizeToolsForAnthropic } from '@shared/anthropic'
import type { Provider } from '@types'
import { APICallError, RetryError } from 'ai'
import { net } from 'electron'
import type { Response } from 'express'
const logger = loggerService.withContext('MessagesService')
@@ -98,11 +100,30 @@ export class MessagesService {
async getClient(provider: Provider, extraHeaders?: Record<string, string | string[]>): Promise<Anthropic> {
// Create Anthropic client for the provider
// Wrap net.fetch to handle compatibility issues:
// 1. net.fetch expects string URLs, not Request objects
// 2. net.fetch doesn't support 'agent' option from Node.js http module
const electronFetch: typeof globalThis.fetch = async (input: URL | RequestInfo, init?: RequestInit) => {
const url = typeof input === 'string' ? input : input instanceof URL ? input.toString() : input.url
// Remove unsupported options for Electron's net.fetch
if (init) {
const initWithAgent = init as RequestInit & { agent?: unknown }
delete initWithAgent.agent
const headers = new Headers(initWithAgent.headers)
if (headers.has('content-length')) {
headers.delete('content-length')
}
initWithAgent.headers = headers
return net.fetch(url, initWithAgent)
}
return net.fetch(url)
}
const context = { fetch: electronFetch }
if (provider.authType === 'oauth') {
const oauthToken = await anthropicService.getValidAccessToken()
return getSdkClient(provider, oauthToken, extraHeaders)
return getSdkClient(provider, oauthToken, extraHeaders, context)
}
return getSdkClient(provider, null, extraHeaders)
return getSdkClient(provider, null, extraHeaders, context)
}
prepareHeaders(headers: Record<string, string | string[] | undefined>): Record<string, string | string[]> {
@@ -127,7 +148,8 @@ export class MessagesService {
createAnthropicRequest(request: MessageCreateParams, provider: Provider, modelId?: string): MessageCreateParams {
const anthropicRequest: MessageCreateParams = {
...request,
stream: !!request.stream
stream: !!request.stream,
tools: sanitizeToolsForAnthropic(request.tools)
}
// Override model if provided
@@ -233,9 +255,71 @@ export class MessagesService {
}
transformError(error: any): { statusCode: number; errorResponse: ErrorResponse } {
let statusCode = 500
let errorType = 'api_error'
let errorMessage = 'Internal server error'
let statusCode: number | undefined = undefined
let errorType: string | undefined = undefined
let errorMessage: string | undefined = undefined
const errorMap: Record<number, string> = {
400: 'invalid_request_error',
401: 'authentication_error',
403: 'forbidden_error',
404: 'not_found_error',
429: 'rate_limit_error',
500: 'internal_server_error'
}
// Handle AI SDK RetryError - extract the last error for better error messages
if (RetryError.isInstance(error)) {
const lastError = error.lastError
// If the last error is an APICallError, extract its details
if (APICallError.isInstance(lastError)) {
statusCode = lastError.statusCode || 502
errorMessage = lastError.message
return {
statusCode,
errorResponse: {
type: 'error',
error: {
type: errorMap[statusCode] || 'api_error',
message: `${error.reason}: ${errorMessage}`,
requestId: lastError.name
}
}
}
}
// Fallback for other retry errors
errorMessage = error.message
statusCode = 502
return {
statusCode,
errorResponse: {
type: 'error',
error: {
type: 'api_error',
message: errorMessage,
requestId: error.name
}
}
}
}
if (APICallError.isInstance(error)) {
statusCode = error.statusCode
errorMessage = error.message
if (statusCode) {
return {
statusCode,
errorResponse: {
type: 'error',
error: {
type: errorMap[statusCode] || 'api_error',
message: errorMessage,
requestId: error.name
}
}
}
}
}
const anthropicStatus = typeof error?.status === 'number' ? error.status : undefined
const anthropicError = error?.error
@@ -277,11 +361,11 @@ export class MessagesService {
typeof errorMessage === 'string' && errorMessage.length > 0 ? errorMessage : 'Internal server error'
return {
statusCode,
statusCode: statusCode ?? 500,
errorResponse: {
type: 'error',
error: {
type: errorType,
type: errorType || 'api_error',
message: safeErrorMessage,
requestId: error?.request_id
}

View File

@@ -1,13 +1,6 @@
import { isEmpty } from 'lodash'
import type { ApiModel, ApiModelsFilter, ApiModelsResponse } from '../../../renderer/src/types/apiModels'
import { loggerService } from '../../services/LoggerService'
import {
getAvailableProviders,
getProviderAnthropicModelChecker,
listAllAvailableModels,
transformModelToOpenAI
} from '../utils'
import { getAvailableProviders, listAllAvailableModels, transformModelToOpenAI } from '../utils'
const logger = loggerService.withContext('ModelsService')
@@ -20,11 +13,12 @@ export class ModelsService {
try {
logger.debug('Getting available models from providers', { filter })
let providers = await getAvailableProviders()
const providers = await getAvailableProviders()
if (filter.providerType === 'anthropic') {
providers = providers.filter((p) => p.type === 'anthropic' || !isEmpty(p.anthropicApiHost?.trim()))
}
// Note: When providerType === 'anthropic', we now return ALL available models
// because the API Server's unified adapter (AiSdkToAnthropicSSE) can convert
// any provider's response to Anthropic SSE format. This enables Claude Code Agent
// to work with OpenAI, Gemini, and other providers transparently.
const models = await listAllAvailableModels(providers)
// Use Map to deduplicate models by their full ID (provider:model_id)
@@ -32,20 +26,11 @@ export class ModelsService {
for (const model of models) {
const provider = providers.find((p) => p.id === model.provider)
// logger.debug(`Processing model ${model.id}`)
if (!provider) {
logger.debug(`Skipping model ${model.id} . Reason: Provider not found.`)
continue
}
if (filter.providerType === 'anthropic') {
const checker = getProviderAnthropicModelChecker(provider.id)
if (!checker(model)) {
logger.debug(`Skipping model ${model.id} from ${model.provider}. Reason: Not an Anthropic model.`)
continue
}
}
const openAIModel = transformModelToOpenAI(model, provider)
const fullModelId = openAIModel.id // This is already in format "provider:model_id"

View File

@@ -0,0 +1,718 @@
import type { AnthropicProviderOptions } from '@ai-sdk/anthropic'
import type { GoogleGenerativeAIProviderOptions } from '@ai-sdk/google'
import type { OpenAIResponsesProviderOptions } from '@ai-sdk/openai'
import type { LanguageModelV2Middleware, LanguageModelV2ToolResultOutput } from '@ai-sdk/provider'
import type { ProviderOptions, ReasoningPart, ToolCallPart, ToolResultPart } from '@ai-sdk/provider-utils'
import type {
ImageBlockParam,
MessageCreateParams,
TextBlockParam,
Tool as AnthropicTool
} from '@anthropic-ai/sdk/resources/messages'
import { type AiPlugin, createExecutor } from '@cherrystudio/ai-core'
import { createProvider as createProviderCore } from '@cherrystudio/ai-core/provider'
import { loggerService } from '@logger'
import { AiSdkToAnthropicSSE, formatSSEDone, formatSSEEvent } from '@main/apiServer/adapters'
import { generateSignature as cherryaiGenerateSignature } from '@main/integration/cherryai'
import anthropicService from '@main/services/AnthropicService'
import copilotService from '@main/services/CopilotService'
import { reduxService } from '@main/services/ReduxService'
import { isGemini3ModelId } from '@shared/middleware'
import {
type AiSdkConfig,
type AiSdkConfigContext,
formatProviderApiHost,
initializeSharedProviders,
isAnthropicProvider,
isGeminiProvider,
isOpenAIProvider,
type ProviderFormatContext,
providerToAiSdkConfig as sharedProviderToAiSdkConfig,
resolveActualProvider
} from '@shared/provider'
import { COPILOT_DEFAULT_HEADERS } from '@shared/provider/constant'
import { defaultAppHeaders } from '@shared/utils'
import type { Provider } from '@types'
import type { ImagePart, JSONValue, ModelMessage, Provider as AiSdkProvider, TextPart, Tool as AiSdkTool } from 'ai'
import { simulateStreamingMiddleware, stepCountIs, tool, wrapLanguageModel, zodSchema } from 'ai'
import { net } from 'electron'
import type { Response } from 'express'
import * as z from 'zod'
import { googleReasoningCache, openRouterReasoningCache } from '../../services/CacheService'
const logger = loggerService.withContext('UnifiedMessagesService')
const MAGIC_STRING = 'skip_thought_signature_validator'
function sanitizeJson(value: unknown): JSONValue {
return JSON.parse(JSON.stringify(value))
}
initializeSharedProviders({
warn: (message) => logger.warn(message),
error: (message, error) => logger.error(message, error)
})
/**
* Configuration for unified message streaming
*/
export interface UnifiedStreamConfig {
response: Response
provider: Provider
modelId: string
params: MessageCreateParams
onError?: (error: unknown) => void
onComplete?: () => void
/**
* Optional AI SDK middlewares to apply
*/
middlewares?: LanguageModelV2Middleware[]
/**
* Optional AI Core plugins to use with the executor
*/
plugins?: AiPlugin[]
}
/**
* Configuration for non-streaming message generation
*/
export interface GenerateUnifiedMessageConfig {
provider: Provider
modelId: string
params: MessageCreateParams
middlewares?: LanguageModelV2Middleware[]
plugins?: AiPlugin[]
}
function getMainProcessFormatContext(): ProviderFormatContext {
const vertexSettings = reduxService.selectSync<{ projectId: string; location: string }>('state.llm.settings.vertexai')
return {
vertex: {
project: vertexSettings?.projectId || 'default-project',
location: vertexSettings?.location || 'us-central1'
}
}
}
const mainProcessSdkContext: AiSdkConfigContext = {
getRotatedApiKey: (provider) => {
const keys = provider.apiKey.split(',').map((k) => k.trim())
return keys[0] || provider.apiKey
},
fetch: net.fetch as typeof globalThis.fetch
}
function getActualProvider(provider: Provider, modelId: string): Provider {
const model = provider.models?.find((m) => m.id === modelId)
if (!model) return provider
return resolveActualProvider(provider, model)
}
function providerToAiSdkConfig(provider: Provider, modelId: string): AiSdkConfig {
const actualProvider = getActualProvider(provider, modelId)
const formattedProvider = formatProviderApiHost(actualProvider, getMainProcessFormatContext())
return sharedProviderToAiSdkConfig(formattedProvider, modelId, mainProcessSdkContext)
}
function convertAnthropicToolResultToAiSdk(
content: string | Array<TextBlockParam | ImageBlockParam>
): LanguageModelV2ToolResultOutput {
if (typeof content === 'string') {
return { type: 'text', value: content }
}
const values: Array<{ type: 'text'; text: string } | { type: 'media'; data: string; mediaType: string }> = []
for (const block of content) {
if (block.type === 'text') {
values.push({ type: 'text', text: block.text })
} else if (block.type === 'image') {
values.push({
type: 'media',
data: block.source.type === 'base64' ? block.source.data : block.source.url,
mediaType: block.source.type === 'base64' ? block.source.media_type : 'image/png'
})
}
}
return { type: 'content', value: values }
}
// Type alias for JSON Schema (compatible with recursive calls)
type JsonSchemaLike = AnthropicTool.InputSchema | Record<string, unknown>
/**
* Convert JSON Schema to Zod schema
* This avoids non-standard fields like input_examples that Anthropic doesn't support
*/
function jsonSchemaToZod(schema: JsonSchemaLike): z.ZodTypeAny {
const s = schema as Record<string, unknown>
const schemaType = s.type as string | string[] | undefined
const enumValues = s.enum as unknown[] | undefined
const description = s.description as string | undefined
// Handle enum first
if (enumValues && Array.isArray(enumValues) && enumValues.length > 0) {
if (enumValues.every((v) => typeof v === 'string')) {
const zodEnum = z.enum(enumValues as [string, ...string[]])
return description ? zodEnum.describe(description) : zodEnum
}
// For non-string enums, use union of literals
const literals = enumValues.map((v) => z.literal(v as string | number | boolean))
if (literals.length === 1) {
return description ? literals[0].describe(description) : literals[0]
}
const zodUnion = z.union(literals as unknown as [z.ZodTypeAny, z.ZodTypeAny, ...z.ZodTypeAny[]])
return description ? zodUnion.describe(description) : zodUnion
}
// Handle union types (type: ["string", "null"])
if (Array.isArray(schemaType)) {
const schemas = schemaType.map((t) => jsonSchemaToZod({ ...s, type: t, enum: undefined }))
if (schemas.length === 1) {
return schemas[0]
}
return z.union(schemas as [z.ZodTypeAny, z.ZodTypeAny, ...z.ZodTypeAny[]])
}
// Handle by type
switch (schemaType) {
case 'string': {
let zodString = z.string()
if (typeof s.minLength === 'number') zodString = zodString.min(s.minLength)
if (typeof s.maxLength === 'number') zodString = zodString.max(s.maxLength)
if (typeof s.pattern === 'string') zodString = zodString.regex(new RegExp(s.pattern))
return description ? zodString.describe(description) : zodString
}
case 'number':
case 'integer': {
let zodNumber = schemaType === 'integer' ? z.number().int() : z.number()
if (typeof s.minimum === 'number') zodNumber = zodNumber.min(s.minimum)
if (typeof s.maximum === 'number') zodNumber = zodNumber.max(s.maximum)
return description ? zodNumber.describe(description) : zodNumber
}
case 'boolean': {
const zodBoolean = z.boolean()
return description ? zodBoolean.describe(description) : zodBoolean
}
case 'null':
return z.null()
case 'array': {
const items = s.items as Record<string, unknown> | undefined
let zodArray = items ? z.array(jsonSchemaToZod(items)) : z.array(z.unknown())
if (typeof s.minItems === 'number') zodArray = zodArray.min(s.minItems)
if (typeof s.maxItems === 'number') zodArray = zodArray.max(s.maxItems)
return description ? zodArray.describe(description) : zodArray
}
case 'object': {
const properties = s.properties as Record<string, Record<string, unknown>> | undefined
const required = (s.required as string[]) || []
// Always use z.object() to ensure "properties" field is present in output schema
// OpenAI requires explicit properties field even for empty objects
const shape: Record<string, z.ZodTypeAny> = {}
if (properties) {
for (const [key, propSchema] of Object.entries(properties)) {
const zodProp = jsonSchemaToZod(propSchema)
shape[key] = required.includes(key) ? zodProp : zodProp.optional()
}
}
const zodObject = z.object(shape)
return description ? zodObject.describe(description) : zodObject
}
default:
// Unknown type, use z.unknown()
return z.unknown()
}
}
function convertAnthropicToolsToAiSdk(tools: MessageCreateParams['tools']): Record<string, AiSdkTool> | undefined {
if (!tools || tools.length === 0) return undefined
const aiSdkTools: Record<string, AiSdkTool> = {}
for (const anthropicTool of tools) {
if (anthropicTool.type === 'bash_20250124') continue
const toolDef = anthropicTool as AnthropicTool
const rawSchema = toolDef.input_schema
const schema = jsonSchemaToZod(rawSchema)
// Use tool() with inputSchema (AI SDK v5 API)
const aiTool = tool({
description: toolDef.description || '',
inputSchema: zodSchema(schema)
})
aiSdkTools[toolDef.name] = aiTool
}
return Object.keys(aiSdkTools).length > 0 ? aiSdkTools : undefined
}
function convertAnthropicToAiMessages(params: MessageCreateParams): ModelMessage[] {
const messages: ModelMessage[] = []
// System message
if (params.system) {
if (typeof params.system === 'string') {
messages.push({ role: 'system', content: params.system })
} else if (Array.isArray(params.system)) {
const systemText = params.system
.filter((block) => block.type === 'text')
.map((block) => block.text)
.join('\n')
if (systemText) {
messages.push({ role: 'system', content: systemText })
}
}
}
const toolCallIdToName = new Map<string, string>()
for (const msg of params.messages) {
if (Array.isArray(msg.content)) {
for (const block of msg.content) {
if (block.type === 'tool_use') {
toolCallIdToName.set(block.id, block.name)
}
}
}
}
// User/assistant messages
for (const msg of params.messages) {
if (typeof msg.content === 'string') {
messages.push({
role: msg.role === 'user' ? 'user' : 'assistant',
content: msg.content
})
} else if (Array.isArray(msg.content)) {
const textParts: TextPart[] = []
const imageParts: ImagePart[] = []
const reasoningParts: ReasoningPart[] = []
const toolCallParts: ToolCallPart[] = []
const toolResultParts: ToolResultPart[] = []
for (const block of msg.content) {
if (block.type === 'text') {
textParts.push({ type: 'text', text: block.text })
} else if (block.type === 'thinking') {
reasoningParts.push({ type: 'reasoning', text: block.thinking })
} else if (block.type === 'redacted_thinking') {
reasoningParts.push({ type: 'reasoning', text: block.data })
} else if (block.type === 'image') {
const source = block.source
if (source.type === 'base64') {
imageParts.push({ type: 'image', image: `data:${source.media_type};base64,${source.data}` })
} else if (source.type === 'url') {
imageParts.push({ type: 'image', image: source.url })
}
} else if (block.type === 'tool_use') {
const options: ProviderOptions = {}
if (isGemini3ModelId(params.model)) {
if (googleReasoningCache.get(`google-${block.name}`)) {
options.google = {
thoughtSignature: MAGIC_STRING
}
} else if (openRouterReasoningCache.get('openrouter')) {
options.openrouter = {
reasoning_details: (sanitizeJson(openRouterReasoningCache.get('openrouter')) as JSONValue[]) || []
}
}
}
toolCallParts.push({
type: 'tool-call',
toolName: block.name,
toolCallId: block.id,
input: block.input,
providerOptions: options
})
} else if (block.type === 'tool_result') {
// Look up toolName from the pre-built map (covers cross-message references)
const toolName = toolCallIdToName.get(block.tool_use_id) || 'unknown'
toolResultParts.push({
type: 'tool-result',
toolCallId: block.tool_use_id,
toolName,
output: block.content ? convertAnthropicToolResultToAiSdk(block.content) : { type: 'text', value: '' }
})
}
}
if (toolResultParts.length > 0) {
messages.push({ role: 'tool', content: [...toolResultParts] })
}
if (msg.role === 'user') {
const userContent = [...textParts, ...imageParts]
if (userContent.length > 0) {
messages.push({ role: 'user', content: userContent })
}
} else {
const assistantContent = [...reasoningParts, ...textParts, ...toolCallParts]
if (assistantContent.length > 0) {
let providerOptions: ProviderOptions | undefined = undefined
if (openRouterReasoningCache.get('openrouter')) {
providerOptions = {
openrouter: {
reasoning_details: (sanitizeJson(openRouterReasoningCache.get('openrouter')) as JSONValue[]) || []
}
}
} else if (isGemini3ModelId(params.model)) {
providerOptions = {
google: {
thoughtSignature: MAGIC_STRING
}
}
}
messages.push({ role: 'assistant', content: assistantContent, providerOptions })
}
}
}
}
return messages
}
interface ExecuteStreamConfig {
provider: Provider
modelId: string
params: MessageCreateParams
middlewares?: LanguageModelV2Middleware[]
plugins?: AiPlugin[]
onEvent?: (event: Parameters<typeof formatSSEEvent>[0]) => void
}
/**
* Create AI SDK provider instance from config
* Similar to renderer's createAiSdkProvider
*/
async function createAiSdkProvider(config: AiSdkConfig): Promise<AiSdkProvider> {
let providerId = config.providerId
// Handle special provider modes (same as renderer)
if (providerId === 'openai' && config.options?.mode === 'chat') {
providerId = 'openai-chat'
} else if (providerId === 'azure' && config.options?.mode === 'responses') {
providerId = 'azure-responses'
} else if (providerId === 'cherryin' && config.options?.mode === 'chat') {
providerId = 'cherryin-chat'
}
const provider = await createProviderCore(providerId, config.options)
return provider
}
/**
* Prepare special provider configuration for providers that need dynamic tokens
* Similar to renderer's prepareSpecialProviderConfig
*/
async function prepareSpecialProviderConfig(provider: Provider, config: AiSdkConfig): Promise<AiSdkConfig> {
switch (provider.id) {
case 'copilot': {
const storedHeaders =
((await reduxService.select('state.copilot.defaultHeaders')) as Record<string, string> | null) ?? {}
const headers: Record<string, string> = {
...COPILOT_DEFAULT_HEADERS,
...storedHeaders
}
try {
const { token } = await copilotService.getToken(null as any, headers)
config.options.apiKey = token
const existingHeaders = (config.options.headers as Record<string, string> | undefined) ?? {}
config.options.headers = {
...headers,
...existingHeaders
}
} catch (error) {
logger.error('Failed to get Copilot token', error as Error)
throw new Error('Failed to get Copilot token. Please re-authorize Copilot.')
}
break
}
case 'anthropic': {
if (provider.authType === 'oauth') {
try {
const oauthToken = await anthropicService.getValidAccessToken()
if (!oauthToken) {
throw new Error('Anthropic OAuth token not available. Please re-authorize.')
}
config.options = {
...config.options,
headers: {
...(config.options.headers ? config.options.headers : {}),
'Content-Type': 'application/json',
'anthropic-version': '2023-06-01',
'anthropic-beta': 'oauth-2025-04-20',
Authorization: `Bearer ${oauthToken}`
},
baseURL: 'https://api.anthropic.com/v1',
apiKey: ''
}
} catch (error) {
logger.error('Failed to get Anthropic OAuth token', error as Error)
throw new Error('Failed to get Anthropic OAuth token. Please re-authorize.')
}
}
break
}
case 'cherryai': {
// Create a signed fetch wrapper for cherryai
const baseFetch = net.fetch as typeof globalThis.fetch
config.options.fetch = async (url: RequestInfo | URL, options?: RequestInit) => {
if (!options?.body) {
return baseFetch(url, options)
}
const signature = cherryaiGenerateSignature({
method: 'POST',
path: '/chat/completions',
query: '',
body: JSON.parse(options.body as string)
})
return baseFetch(url, {
...options,
headers: {
...(options.headers as Record<string, string>),
...signature
}
})
}
break
}
}
return config
}
function mapAnthropicThinkToAISdkProviderOptions(
provider: Provider,
config: MessageCreateParams['thinking']
): ProviderOptions | undefined {
if (!config) return undefined
if (isAnthropicProvider(provider)) {
return {
anthropic: {
...mapToAnthropicProviderOptions(config)
}
}
}
if (isGeminiProvider(provider)) {
return {
google: {
...mapToGeminiProviderOptions(config)
}
}
}
if (isOpenAIProvider(provider)) {
return {
openai: {
...mapToOpenAIProviderOptions(config)
}
}
}
return undefined
}
function mapToAnthropicProviderOptions(config: NonNullable<MessageCreateParams['thinking']>): AnthropicProviderOptions {
return {
thinking: {
type: config.type,
budgetTokens: config.type === 'enabled' ? config.budget_tokens : undefined
}
}
}
function mapToGeminiProviderOptions(
config: NonNullable<MessageCreateParams['thinking']>
): GoogleGenerativeAIProviderOptions {
return {
thinkingConfig: {
thinkingBudget: config.type === 'enabled' ? config.budget_tokens : -1,
includeThoughts: config.type === 'enabled'
}
}
}
function mapToOpenAIProviderOptions(
config: NonNullable<MessageCreateParams['thinking']>
): OpenAIResponsesProviderOptions {
return {
reasoningEffort: config.type === 'enabled' ? 'high' : 'none'
}
}
/**
* Core stream execution function - single source of truth for AI SDK calls
*/
async function executeStream(config: ExecuteStreamConfig): Promise<AiSdkToAnthropicSSE> {
const { provider, modelId, params, middlewares = [], plugins = [], onEvent } = config
// Convert provider config to AI SDK config
let sdkConfig = providerToAiSdkConfig(provider, modelId)
// Prepare special provider config (Copilot, Anthropic OAuth, etc.)
sdkConfig = await prepareSpecialProviderConfig(provider, sdkConfig)
// Create provider instance and get language model
const aiSdkProvider = await createAiSdkProvider(sdkConfig)
const baseModel = aiSdkProvider.languageModel(modelId)
// Apply middlewares if present
const model =
middlewares.length > 0 && typeof baseModel === 'object'
? (wrapLanguageModel({ model: baseModel, middleware: middlewares }) as typeof baseModel)
: baseModel
// Create executor with plugins
const executor = createExecutor(sdkConfig.providerId, sdkConfig.options, plugins)
// Convert messages and tools
const coreMessages = convertAnthropicToAiMessages(params)
const tools = convertAnthropicToolsToAiSdk(params.tools)
// Create the adapter
const adapter = new AiSdkToAnthropicSSE({
model: `${provider.id}:${modelId}`,
onEvent: onEvent || (() => {})
})
// Execute stream - pass model object instead of string
const result = await executor.streamText({
model, // Now passing LanguageModel object, not string
messages: coreMessages,
// FIXME: Claude Code传入的maxToken会超出有些模型限制需做特殊处理可能在v2好修复一点现在维护的成本有点高
// 已知: 豆包
maxOutputTokens: params.max_tokens,
temperature: params.temperature,
topP: params.top_p,
topK: params.top_k,
stopSequences: params.stop_sequences,
stopWhen: stepCountIs(100),
headers: defaultAppHeaders(),
tools,
providerOptions: mapAnthropicThinkToAISdkProviderOptions(provider, params.thinking)
})
// Process the stream through the adapter
await adapter.processStream(result.fullStream)
return adapter
}
/**
* Stream a message request using AI SDK executor and convert to Anthropic SSE format
*/
export async function streamUnifiedMessages(config: UnifiedStreamConfig): Promise<void> {
const { response, provider, modelId, params, onError, onComplete, middlewares = [], plugins = [] } = config
logger.info('Starting unified message stream', {
providerId: provider.id,
providerType: provider.type,
modelId,
stream: params.stream,
middlewareCount: middlewares.length,
pluginCount: plugins.length
})
try {
response.setHeader('Content-Type', 'text/event-stream')
response.setHeader('Cache-Control', 'no-cache')
response.setHeader('Connection', 'keep-alive')
response.setHeader('X-Accel-Buffering', 'no')
await executeStream({
provider,
modelId,
params,
middlewares,
plugins,
onEvent: (event) => {
logger.silly('Streaming event', { eventType: event.type })
const sseData = formatSSEEvent(event)
response.write(sseData)
}
})
// Send done marker
response.write(formatSSEDone())
response.end()
logger.info('Unified message stream completed', { providerId: provider.id, modelId })
onComplete?.()
} catch (error) {
logger.error('Error in unified message stream', error as Error, { providerId: provider.id, modelId })
onError?.(error)
throw error
}
}
/**
* Generate a non-streaming message response
*
* Uses simulateStreamingMiddleware to reuse the same streaming logic,
* similar to renderer's ModernAiProvider pattern.
*/
export async function generateUnifiedMessage(
providerOrConfig: Provider | GenerateUnifiedMessageConfig,
modelId?: string,
params?: MessageCreateParams
): Promise<ReturnType<typeof AiSdkToAnthropicSSE.prototype.buildNonStreamingResponse>> {
// Support both old signature and new config-based signature
let config: GenerateUnifiedMessageConfig
if ('provider' in providerOrConfig && 'modelId' in providerOrConfig && 'params' in providerOrConfig) {
config = providerOrConfig
} else {
config = {
provider: providerOrConfig as Provider,
modelId: modelId!,
params: params!
}
}
const { provider, middlewares = [], plugins = [] } = config
logger.info('Starting unified message generation', {
providerId: provider.id,
providerType: provider.type,
modelId: config.modelId,
middlewareCount: middlewares.length,
pluginCount: plugins.length
})
try {
// Add simulateStreamingMiddleware to reuse streaming logic for non-streaming
const allMiddlewares = [simulateStreamingMiddleware(), ...middlewares]
const adapter = await executeStream({
provider,
modelId: config.modelId,
params: config.params,
middlewares: allMiddlewares,
plugins
})
const finalResponse = adapter.buildNonStreamingResponse()
logger.info('Unified message generation completed', {
providerId: provider.id,
modelId: config.modelId
})
return finalResponse
} catch (error) {
logger.error('Error in unified message generation', error as Error, {
providerId: provider.id,
modelId: config.modelId
})
throw error
}
}
export default {
streamUnifiedMessages,
generateUnifiedMessage
}

View File

@@ -1,6 +1,7 @@
import { CacheService } from '@main/services/CacheService'
import { loggerService } from '@main/services/LoggerService'
import { reduxService } from '@main/services/ReduxService'
import { isPpioAnthropicCompatibleModel, isSiliconAnthropicCompatibleModel } from '@shared/config/providers'
import type { ApiModel, Model, Provider } from '@types'
const logger = loggerService.withContext('ApiServerUtils')
@@ -27,10 +28,9 @@ export async function getAvailableProviders(): Promise<Provider[]> {
return []
}
// Support OpenAI and Anthropic type providers for API server
const supportedProviders = providers.filter(
(p: Provider) => p.enabled && (p.type === 'openai' || p.type === 'anthropic')
)
// Support all provider types that AI SDK can handle
// The unified-messages service uses AI SDK which supports many providers
const supportedProviders = providers.filter((p: Provider) => p.enabled)
// Cache the filtered results
CacheService.set(PROVIDERS_CACHE_KEY, supportedProviders, PROVIDERS_CACHE_TTL)
@@ -159,7 +159,7 @@ export async function validateModelId(model: string): Promise<{
valid: false,
error: {
type: 'provider_not_found',
message: `Provider '${providerId}' not found, not enabled, or not supported. Only OpenAI providers are currently supported.`,
message: `Provider '${providerId}' not found or not enabled.`,
code: 'provider_not_found'
}
}
@@ -261,14 +261,8 @@ export function validateProvider(provider: Provider): boolean {
return false
}
// Support OpenAI and Anthropic type providers
if (provider.type !== 'openai' && provider.type !== 'anthropic') {
logger.debug('Provider type not supported', {
providerId: provider.id,
providerType: provider.type
})
return false
}
// AI SDK supports many provider types, no longer need to filter by type
// The unified-messages service handles all supported types
return true
} catch (error: any) {
@@ -287,8 +281,41 @@ export const getProviderAnthropicModelChecker = (providerId: string): ((m: Model
return (m: Model) => m.endpoint_type === 'anthropic'
case 'aihubmix':
return (m: Model) => m.id.includes('claude')
case 'silicon':
return (m: Model) => isSiliconAnthropicCompatibleModel(m.id)
case 'ppio':
return (m: Model) => isPpioAnthropicCompatibleModel(m.id)
default:
// allow all models when checker not configured
return () => true
}
}
/**
* Check if a specific model is compatible with Anthropic API for a given provider.
*
* This is used for fine-grained routing decisions at the model level.
* For aggregated providers (like Silicon), only certain models support the Anthropic API endpoint.
*
* @param provider - The provider to check
* @param modelId - The model ID to check (without provider prefix)
* @returns true if the model supports Anthropic API endpoint
*/
export function isModelAnthropicCompatible(provider: Provider, modelId: string): boolean {
const checker = getProviderAnthropicModelChecker(provider.id)
const model = provider.models?.find((m) => m.id === modelId)
if (model) {
return checker(model)
}
const minimalModel: Model = {
id: modelId,
name: modelId,
provider: provider.id,
group: ''
}
return checker(minimalModel)
}

View File

@@ -1,9 +1,19 @@
import type { ReasoningDetailUnion } from '@main/apiServer/adapters/openrouter'
interface CacheItem<T> {
data: T
timestamp: number
duration: number
}
/**
* Interface for reasoning cache
*/
export interface IReasoningCache<T> {
set(key: string, value: T): void
get(key: string): T | undefined
}
export class CacheService {
private static cache: Map<string, CacheItem<any>> = new Map()
@@ -72,3 +82,14 @@ export class CacheService {
return true
}
}
// Singleton cache instances using CacheService
export const googleReasoningCache: IReasoningCache<string> = {
set: (key, value) => CacheService.set(`google-reasoning:${key}`, value, 30 * 60 * 1000),
get: (key) => CacheService.get(`google-reasoning:${key}`) || undefined
}
export const openRouterReasoningCache: IReasoningCache<ReasoningDetailUnion[]> = {
set: (key, value) => CacheService.set(`openrouter-reasoning:${key}`, value, 30 * 60 * 1000),
get: (key) => CacheService.get(`openrouter-reasoning:${key}`) || undefined
}

View File

@@ -548,6 +548,17 @@ class CodeToolsService {
logger.debug(`Environment variables:`, Object.keys(env))
logger.debug(`Options:`, options)
// Validate directory exists before proceeding
if (!directory || !fs.existsSync(directory)) {
const errorMessage = `Directory does not exist: ${directory}`
logger.error(errorMessage)
return {
success: false,
message: errorMessage,
command: ''
}
}
const packageName = await this.getPackageName(cliTool)
const bunPath = await this.getBunPath()
const executableName = await this.getCliExecutableName(cliTool)
@@ -709,6 +720,7 @@ class CodeToolsService {
// Build bat file content, including debug information
const batContent = [
'@echo off',
'chcp 65001 >nul 2>&1', // Switch to UTF-8 code page for international path support
`title ${cliTool} - Cherry Studio`, // Set window title in bat file
'echo ================================================',
'echo Cherry Studio CLI Tool Launcher',

View File

@@ -620,7 +620,7 @@ class McpService {
tools.map((tool: SDKTool) => {
const serverTool: MCPTool = {
...tool,
id: buildFunctionCallToolName(server.name, tool.name),
id: buildFunctionCallToolName(server.name, tool.name, server.id),
serverId: server.id,
serverName: server.name,
type: 'mcp'

View File

@@ -87,6 +87,7 @@ export class ClaudeStreamState {
private pendingUsage: PendingUsageState = {}
private pendingToolCalls = new Map<string, PendingToolCall>()
private stepActive = false
private _streamFinished = false
constructor(options: ClaudeStreamStateOptions) {
this.logger = loggerService.withContext('ClaudeStreamState')
@@ -289,6 +290,16 @@ export class ClaudeStreamState {
getNamespacedToolCallId(rawToolCallId: string): string {
return buildNamespacedToolCallId(this.agentSessionId, rawToolCallId)
}
/** Marks the stream as finished (either completed or errored). */
markFinished(): void {
this._streamFinished = true
}
/** Returns true if the stream has already emitted a terminal event. */
isFinished(): boolean {
return this._streamFinished
}
}
export type { PendingToolCall }

View File

@@ -1,6 +1,7 @@
// src/main/services/agents/services/claudecode/index.ts
import { EventEmitter } from 'node:events'
import { createRequire } from 'node:module'
import path from 'node:path'
import type {
CanUseTool,
@@ -84,18 +85,14 @@ class ClaudeCodeService implements AgentServiceInterface {
})
return aiStream
}
if (
(modelInfo.provider?.type !== 'anthropic' &&
(modelInfo.provider?.anthropicApiHost === undefined || modelInfo.provider.anthropicApiHost.trim() === '')) ||
modelInfo.provider.apiKey === ''
) {
logger.error('Anthropic provider configuration is missing', {
modelInfo
})
// Validate provider has required configuration
// Note: We no longer restrict to anthropic type only - the API Server's unified adapter
// handles format conversion for any provider type (OpenAI, Gemini, etc.)
if (!modelInfo.provider?.apiKey) {
logger.error('Provider API key is missing', { modelInfo })
aiStream.emit('data', {
type: 'error',
error: new Error(`Invalid provider type '${modelInfo.provider?.type}'. Expected 'anthropic' provider type.`)
error: new Error(`Provider '${modelInfo.provider?.id}' is missing API key configuration.`)
})
return aiStream
}
@@ -106,22 +103,25 @@ class ClaudeCodeService implements AgentServiceInterface {
Object.entries(loginShellEnv).filter(([key]) => !key.toLowerCase().endsWith('_proxy'))
) as Record<string, string>
// Route through local API Server which handles format conversion via unified adapter
// This enables Claude Code Agent to work with any provider (OpenAI, Gemini, etc.)
// The API Server converts AI SDK responses to Anthropic SSE format transparently
const env = {
...loginShellEnvWithoutProxies,
// TODO: fix the proxy api server
// ANTHROPIC_API_KEY: apiConfig.apiKey,
// ANTHROPIC_AUTH_TOKEN: apiConfig.apiKey,
// ANTHROPIC_BASE_URL: `http://${apiConfig.host}:${apiConfig.port}/${modelInfo.provider.id}`,
ANTHROPIC_API_KEY: modelInfo.provider.apiKey,
ANTHROPIC_AUTH_TOKEN: modelInfo.provider.apiKey,
ANTHROPIC_BASE_URL: modelInfo.provider.anthropicApiHost?.trim() || modelInfo.provider.apiHost,
ANTHROPIC_API_KEY: apiConfig.apiKey,
ANTHROPIC_AUTH_TOKEN: apiConfig.apiKey,
ANTHROPIC_BASE_URL: `http://${apiConfig.host}:${apiConfig.port}/${modelInfo.provider.id}`,
ANTHROPIC_MODEL: modelInfo.modelId,
ANTHROPIC_DEFAULT_OPUS_MODEL: modelInfo.modelId,
ANTHROPIC_DEFAULT_SONNET_MODEL: modelInfo.modelId,
// TODO: support set small model in UI
ANTHROPIC_DEFAULT_HAIKU_MODEL: modelInfo.modelId,
ELECTRON_RUN_AS_NODE: '1',
ELECTRON_NO_ATTACH_CONSOLE: '1'
ELECTRON_NO_ATTACH_CONSOLE: '1',
// Set CLAUDE_CONFIG_DIR to app's userData directory to avoid path encoding issues
// on Windows when the username contains non-ASCII characters (e.g., Chinese characters)
// This prevents the SDK from using the user's home directory which may have encoding problems
CLAUDE_CONFIG_DIR: path.join(app.getPath('userData'), '.claude')
}
const errorChunks: string[] = []
@@ -534,6 +534,19 @@ class ClaudeCodeService implements AgentServiceInterface {
return
}
// Skip emitting error if stream already finished (error was handled via result message)
if (streamState.isFinished()) {
logger.debug('SDK process exited after stream finished, skipping duplicate error event', {
duration,
error: errorObj instanceof Error ? { name: errorObj.name, message: errorObj.message } : String(errorObj)
})
// Still emit complete to signal stream end
stream.emit('data', {
type: 'complete'
})
return
}
errorChunks.push(errorObj instanceof Error ? errorObj.message : String(errorObj))
const errorMessage = errorChunks.join('\n\n')
logger.error('SDK query failed', {

View File

@@ -121,7 +121,7 @@ export function transformSDKMessageToStreamParts(sdkMessage: SDKMessage, state:
case 'system':
return handleSystemMessage(sdkMessage)
case 'result':
return handleResultMessage(sdkMessage)
return handleResultMessage(sdkMessage, state)
default:
logger.warn('Unknown SDKMessage type', { type: (sdkMessage as any).type })
return []
@@ -193,6 +193,30 @@ function handleAssistantMessage(
}
break
}
case 'thinking':
case 'redacted_thinking': {
const thinkingText = block.type === 'thinking' ? block.thinking : block.data
if (thinkingText) {
const id = generateMessageId()
chunks.push({
type: 'reasoning-start',
id,
providerMetadata
})
chunks.push({
type: 'reasoning-delta',
id,
text: thinkingText,
providerMetadata
})
chunks.push({
type: 'reasoning-end',
id,
providerMetadata
})
}
break
}
case 'tool_use':
handleAssistantToolUse(block as ToolUseContent, providerMetadata, state, chunks)
break
@@ -445,7 +469,11 @@ function handleStreamEvent(
case 'content_block_stop': {
const block = state.closeBlock(event.index)
if (!block) {
logger.warn('Received content_block_stop for unknown index', { index: event.index })
// Some providers (e.g., Gemini) send content via assistant message before stream events,
// so the block may not exist in state. This is expected behavior, not an error.
logger.debug('Received content_block_stop for unknown index (may be from non-streaming content)', {
index: event.index
})
break
}
@@ -679,7 +707,13 @@ function handleSystemMessage(message: Extract<SDKMessage, { type: 'system' }>):
* Successful runs yield a `finish` frame with aggregated usage metrics, while
* failures are surfaced as `error` frames.
*/
function handleResultMessage(message: Extract<SDKMessage, { type: 'result' }>): AgentStreamPart[] {
function handleResultMessage(
message: Extract<SDKMessage, { type: 'result' }>,
state: ClaudeStreamState
): AgentStreamPart[] {
// Mark stream as finished to prevent duplicate error events when SDK process exits
state.markFinished()
const chunks: AgentStreamPart[] = []
let usage: LanguageModelUsage | undefined
@@ -691,26 +725,33 @@ function handleResultMessage(message: Extract<SDKMessage, { type: 'result' }>):
}
}
if (message.subtype === 'success') {
chunks.push({
type: 'finish',
totalUsage: usage ?? emptyUsage,
finishReason: mapClaudeCodeFinishReason(message.subtype),
providerMetadata: {
...sdkMessageToProviderMetadata(message),
usage: message.usage,
durationMs: message.duration_ms,
costUsd: message.total_cost_usd,
raw: message
}
} as AgentStreamPart)
} else {
chunks.push({
type: 'finish',
totalUsage: usage ?? emptyUsage,
finishReason: mapClaudeCodeFinishReason(message.subtype),
providerMetadata: {
...sdkMessageToProviderMetadata(message),
usage: message.usage,
durationMs: message.duration_ms,
costUsd: message.total_cost_usd,
raw: message
}
} as AgentStreamPart)
if (message.subtype !== 'success') {
chunks.push({
type: 'error',
error: {
message: `${message.subtype}: Process failed after ${message.num_turns} turns`
}
} as AgentStreamPart)
} else {
if (message.is_error) {
const errorMatch = message.result.match(/\{.*\}/)
if (errorMatch) {
const errorDetail = JSON.parse(errorMatch[0])
chunks.push(errorDetail)
}
}
}
return chunks
}

View File

@@ -0,0 +1,196 @@
import { describe, expect, it } from 'vitest'
import { buildFunctionCallToolName } from '../mcp'
describe('buildFunctionCallToolName', () => {
describe('basic functionality', () => {
it('should combine server name and tool name', () => {
const result = buildFunctionCallToolName('github', 'search_issues')
expect(result).toContain('github')
expect(result).toContain('search')
})
it('should sanitize names by replacing dashes with underscores', () => {
const result = buildFunctionCallToolName('my-server', 'my-tool')
// Input dashes are replaced, but the separator between server and tool is a dash
expect(result).toBe('my_serv-my_tool')
expect(result).toContain('_')
})
it('should handle empty server names gracefully', () => {
const result = buildFunctionCallToolName('', 'tool')
expect(result).toBeTruthy()
})
})
describe('uniqueness with serverId', () => {
it('should generate different IDs for same server name but different serverIds', () => {
const serverId1 = 'server-id-123456'
const serverId2 = 'server-id-789012'
const serverName = 'github'
const toolName = 'search_repos'
const result1 = buildFunctionCallToolName(serverName, toolName, serverId1)
const result2 = buildFunctionCallToolName(serverName, toolName, serverId2)
expect(result1).not.toBe(result2)
expect(result1).toContain('123456')
expect(result2).toContain('789012')
})
it('should generate same ID when serverId is not provided', () => {
const serverName = 'github'
const toolName = 'search_repos'
const result1 = buildFunctionCallToolName(serverName, toolName)
const result2 = buildFunctionCallToolName(serverName, toolName)
expect(result1).toBe(result2)
})
it('should include serverId suffix when provided', () => {
const serverId = 'abc123def456'
const result = buildFunctionCallToolName('server', 'tool', serverId)
// Should include last 6 chars of serverId
expect(result).toContain('ef456')
})
})
describe('character sanitization', () => {
it('should replace invalid characters with underscores', () => {
const result = buildFunctionCallToolName('test@server', 'tool#name')
expect(result).not.toMatch(/[@#]/)
expect(result).toMatch(/^[a-zA-Z0-9_-]+$/)
})
it('should ensure name starts with a letter', () => {
const result = buildFunctionCallToolName('123server', '456tool')
expect(result).toMatch(/^[a-zA-Z]/)
})
it('should handle consecutive underscores/dashes', () => {
const result = buildFunctionCallToolName('my--server', 'my__tool')
expect(result).not.toMatch(/[_-]{2,}/)
})
})
describe('length constraints', () => {
it('should truncate names longer than 63 characters', () => {
const longServerName = 'a'.repeat(50)
const longToolName = 'b'.repeat(50)
const result = buildFunctionCallToolName(longServerName, longToolName, 'id123456')
expect(result.length).toBeLessThanOrEqual(63)
})
it('should not end with underscore or dash after truncation', () => {
const longServerName = 'a'.repeat(50)
const longToolName = 'b'.repeat(50)
const result = buildFunctionCallToolName(longServerName, longToolName, 'id123456')
expect(result).not.toMatch(/[_-]$/)
})
it('should preserve serverId suffix even with long server/tool names', () => {
const longServerName = 'a'.repeat(50)
const longToolName = 'b'.repeat(50)
const serverId = 'server-id-xyz789'
const result = buildFunctionCallToolName(longServerName, longToolName, serverId)
// The suffix should be preserved and not truncated
expect(result).toContain('xyz789')
expect(result.length).toBeLessThanOrEqual(63)
})
it('should ensure two long-named servers with different IDs produce different results', () => {
const longServerName = 'a'.repeat(50)
const longToolName = 'b'.repeat(50)
const serverId1 = 'server-id-abc123'
const serverId2 = 'server-id-def456'
const result1 = buildFunctionCallToolName(longServerName, longToolName, serverId1)
const result2 = buildFunctionCallToolName(longServerName, longToolName, serverId2)
// Both should be within limit
expect(result1.length).toBeLessThanOrEqual(63)
expect(result2.length).toBeLessThanOrEqual(63)
// They should be different due to preserved suffix
expect(result1).not.toBe(result2)
})
})
describe('edge cases with serverId', () => {
it('should handle serverId with only non-alphanumeric characters', () => {
const serverId = '------' // All dashes
const result = buildFunctionCallToolName('server', 'tool', serverId)
// Should still produce a valid unique suffix via fallback hash
expect(result).toBeTruthy()
expect(result.length).toBeLessThanOrEqual(63)
expect(result).toMatch(/^[a-zA-Z][a-zA-Z0-9_-]*$/)
// Should have a suffix (underscore followed by something)
expect(result).toMatch(/_[a-z0-9]+$/)
})
it('should produce different results for different non-alphanumeric serverIds', () => {
const serverId1 = '------'
const serverId2 = '!!!!!!'
const result1 = buildFunctionCallToolName('server', 'tool', serverId1)
const result2 = buildFunctionCallToolName('server', 'tool', serverId2)
// Should be different because the hash fallback produces different values
expect(result1).not.toBe(result2)
})
it('should handle empty string serverId differently from undefined', () => {
const resultWithEmpty = buildFunctionCallToolName('server', 'tool', '')
const resultWithUndefined = buildFunctionCallToolName('server', 'tool', undefined)
// Empty string is falsy, so both should behave the same (no suffix)
expect(resultWithEmpty).toBe(resultWithUndefined)
})
it('should handle serverId with mixed alphanumeric and special chars', () => {
const serverId = 'ab@#cd' // Mixed chars, last 6 chars contain some alphanumeric
const result = buildFunctionCallToolName('server', 'tool', serverId)
// Should extract alphanumeric chars: 'abcd' from 'ab@#cd'
expect(result).toContain('abcd')
})
})
describe('real-world scenarios', () => {
it('should handle GitHub MCP server instances correctly', () => {
const serverName = 'github'
const toolName = 'search_repositories'
const githubComId = 'server-github-com-abc123'
const gheId = 'server-ghe-internal-xyz789'
const tool1 = buildFunctionCallToolName(serverName, toolName, githubComId)
const tool2 = buildFunctionCallToolName(serverName, toolName, gheId)
// Should be different
expect(tool1).not.toBe(tool2)
// Both should be valid identifiers
expect(tool1).toMatch(/^[a-zA-Z][a-zA-Z0-9_-]*$/)
expect(tool2).toMatch(/^[a-zA-Z][a-zA-Z0-9_-]*$/)
// Both should be <= 63 chars
expect(tool1.length).toBeLessThanOrEqual(63)
expect(tool2.length).toBeLessThanOrEqual(63)
})
it('should handle tool names that already include server name prefix', () => {
const result = buildFunctionCallToolName('github', 'github_search_repos')
expect(result).toBeTruthy()
// Should not double the server name
expect(result.split('github').length - 1).toBeLessThanOrEqual(2)
})
})
})

View File

@@ -1,7 +1,25 @@
export function buildFunctionCallToolName(serverName: string, toolName: string) {
export function buildFunctionCallToolName(serverName: string, toolName: string, serverId?: string) {
const sanitizedServer = serverName.trim().replace(/-/g, '_')
const sanitizedTool = toolName.trim().replace(/-/g, '_')
// Calculate suffix first to reserve space for it
// Suffix format: "_" + 6 alphanumeric chars = 7 chars total
let serverIdSuffix = ''
if (serverId) {
// Take the last 6 characters of the serverId for brevity
serverIdSuffix = serverId.slice(-6).replace(/[^a-zA-Z0-9]/g, '')
// Fallback: if suffix becomes empty (all non-alphanumeric chars), use a simple hash
if (!serverIdSuffix) {
const hash = serverId.split('').reduce((acc, char) => acc + char.charCodeAt(0), 0)
serverIdSuffix = hash.toString(36).slice(-6) || 'x'
}
}
// Reserve space for suffix when calculating max base name length
const SUFFIX_LENGTH = serverIdSuffix ? serverIdSuffix.length + 1 : 0 // +1 for underscore
const MAX_BASE_LENGTH = 63 - SUFFIX_LENGTH
// Combine server name and tool name
let name = sanitizedTool
if (!sanitizedTool.includes(sanitizedServer.slice(0, 7))) {
@@ -20,9 +38,9 @@ export function buildFunctionCallToolName(serverName: string, toolName: string)
// Remove consecutive underscores/dashes (optional improvement)
name = name.replace(/[_-]{2,}/g, '_')
// Truncate to 63 characters maximum
if (name.length > 63) {
name = name.slice(0, 63)
// Truncate base name BEFORE adding suffix to ensure suffix is never cut off
if (name.length > MAX_BASE_LENGTH) {
name = name.slice(0, MAX_BASE_LENGTH)
}
// Handle edge case: ensure we still have a valid name if truncation left invalid chars at edges
@@ -30,5 +48,10 @@ export function buildFunctionCallToolName(serverName: string, toolName: string)
name = name.slice(0, -1)
}
// Now append the suffix - it will always fit within 63 chars
if (serverIdSuffix) {
name = `${name}_${serverIdSuffix}`
}
return name
}

View File

@@ -212,8 +212,9 @@ export class ToolCallChunkHandler {
description: toolName,
type: 'builtin'
} as BaseTool
} else if ((mcpTool = this.mcpTools.find((t) => t.name === toolName) as MCPTool)) {
} else if ((mcpTool = this.mcpTools.find((t) => t.id === toolName) as MCPTool)) {
// 如果是客户端执行的 MCP 工具,沿用现有逻辑
// toolName is mcpTool.id (registered with id as key in convertMcpToolsToAiSdkTools)
logger.info(`[ToolCallChunkHandler] Handling client-side MCP tool: ${toolName}`)
// mcpTool = this.mcpTools.find((t) => t.name === toolName) as MCPTool
// if (!mcpTool) {

Some files were not shown because too many files have changed in this diff Show More