Improve robustness of tool call handling in OpenAI completions and agent tool loop by avoiding premature filtering and surfacing clearer errors when tools are missing.
* Refactor tool call argument handling in openai_source.py
* Improve error logging for missing tools
Log available tools when a specified tool is not found.
* feat: add MiniMax Token Plan provider with hardcoded model list (fix#7585)
- Add new provider 'minimax_token_plan' for MiniMax Token Plan users
- Inherit ProviderAnthropic to reuse all chat/completion logic
- Hardcode api_base to https://api.minimaxi.com/anthropic
- get_models() returns hardcoded list: MiniMax-M2.7, M2.5, M2.1, M2
- Highspeed models excluded (require premium tier)
- Reason for hardcoding: Token Plan API does not expose /models endpoint
- Fixes: https://github.com/AstrBotDevs/AstrBot/issues/7585
* fix: remove api_base from config template and add model validation
- Remove api_base from default_config_tmpl (always overridden, misleading)
- Add model validation against MINIMAX_TOKEN_PLAN_MODELS
- Raise clear ValueError if user configures an unsupported model
Addressed Sourcery AI review comments.
* fix: use custom_headers for Bearer token auth instead of auth_header
MiniMax Token Plan requires Authorization: Bearer <token> header.
Use custom_headers to inject the correct auth header instead of
the non-functional auth_header key.
Addressed Gemini Code Assist review comment.
* fix: update MiniMax Token Plan provider adapter and documentation to English
* feat: add MiniMax Token Plan configuration and icon support
* feat: remove default configuration template from MiniMax Token Plan provider adapter
---------
Co-authored-by: Soulter <905617992@qq.com>
* feat: Refactor astrbot builtin tool management
- Introduced a new registry for builtin tools to streamline their management.
- Added `SendMessageToUserTool`, `KnowledgeBaseQueryTool`, and various web search tools as builtin tools.
- Updated `FunctionToolManager` to cache and retrieve builtin tools efficiently.
- Modified `CronJobManager` to utilize the new `SendMessageToUserTool`.
- Enhanced the dashboard to display readonly status for builtin tools and prevent toggling their state.
- Added tests for builtin tool injection and retrieval to ensure proper functionality.
* fix: escape file path in shell command to prevent injection vulnerabilities
* fix: skip FunctionCallingConfig when only native tools are present
When native tools (google_search, url_context) are enabled without any
function_declarations, _prepare_query_config was still creating a
FunctionCallingConfig, which makes Gemini API return 400 INVALID_ARGUMENT.
Now we only set tool_config when tool_list actually contains
function_declarations.
Fixes#7406
* style: ruff format
* feat: add Rerank API support for NVIDIA NIM
- Add Rerank API support for NVIDIA NIM
- Add related i18n support in en-US zh-CN
* chore: format code
* fix: replace illegal characters
Replace illegal characters when building request model path.
* fix: refactor client initialization method
* fix: enhance response parsing
* docs: add comment for model_path process
* docs: add russia translation
* feat: update AddNewProvider component to support current provider type and enhance provider icon mapping
---------
Co-authored-by: Soulter <905617992@qq.com>
* feat: add audio input support across providers and chatui recording issue fix
- Introduced audio_urls parameter in Provider class and related methods to handle audio input.
- Updated ProviderAnthropic, ProviderGoogleGenAI, and ProviderOpenAIOfficial to process audio URLs.
- Enhanced media_utils with functions to ensure audio format compatibility and detect audio types.
- Modified dashboard components to display audio input support and handle audio attachments in messages.
- Updated localization files to include audio as a supported modality.
- Added new icons for audio input in the dashboard UI.
* feat: enhance audio handling with temporary file cleanup and format support
* feat: track temporary local files for converted audio components
* fix: update image placeholder in prompt from "[图片]" to "[Image]"
* feat(provider/vllm_rerank): add configurable rerank_api_suffix option
Add rerank_api_suffix config option to the VLLM Rerank provider so
users can control the API URL path suffix instead of having /v1/rerank
hardcoded.
- Default value is /v1/rerank (preserves existing behavior)
- Users can set it to empty string to disable auto-append
- Handles suffix without leading slash by auto-adding one
- Schema, default config, and i18n metadata all updated
Issue: Fixes#7238
* fix(provider/vllm_rerank): handle null suffix and improve hint descriptions
- Add explicit None check for rerank_api_suffix (fixes HIGH from Gemini)
- Update rerank_api_base hint to describe actual behavior without
mentioning specific provider options (fixes 3x MEDIUM from Gemini)
- Add ru-RU i18n for rerank_api_suffix (fixes P2 from Codex)
Co-authored-by: gemini-code-assist[bot]
Co-authored-by: chatgpt-codex-connector[bot]
---------
Co-authored-by: LehaoLin <linlehao@cuhk.edu.cn>
Gemini 3 models return thinking parts (part.thought=True) alongside the
actual response text. _process_content_parts was including these thinking
parts in the message chain sent to the user, effectively leaking internal
reasoning into the output. On platforms that split long messages (e.g.
aiocqhttp with realtime segmenting), this caused duplicate or triple
replies since the thinking text often mirrors the actual response.
The streaming path already handled this correctly via chunk.text which
skips thinking parts, but the non-streaming path and the final-chunk
processing in streaming both went through _process_content_parts.
Also switch the Gemini 3 model name matching from an exhaustive list to
prefix matching (gemini-3- / gemini-3.) so new variants like gemini-3.1
get proper thinkingLevel config without code changes.
Fixes#7183
Gemini API requires function_response to be a google.protobuf.Struct
(JSON object). When tool results are plain text strings, the API
returns 400 Invalid argument. Detect non-JSON tool content for Gemini
models and wrap it in {"result": content} before sending.
Fixes#7134
* fix: filter empty assistant messages to prevent 400 error on strict APIs
Some OpenAI-compatible APIs (e.g., Moonshot) reject requests with
empty content in assistant messages when no tool_calls are present.
This fix cleans up the messages payload before sending to avoid
'message at position X must not be empty' errors.
Closes related issue with fallback provider behavior.
* test(openai): add tests for empty assistant message filtering
* refactor(openai): simplify empty assistant message filtering logic
* style: format code
---------
Co-authored-by: RC-CHN <1051989940@qq.com>
* fix: support both old and new Bailian Rerank API response formats
The new compatible API (compatible-api/v1/reranks) returns results at
the top level as data.results, while the old API returns them nested
under data.output.results. The parser only checked the old path,
causing qwen3-rerank to always report empty results.
Fixes#7161
* Update astrbot/core/provider/sources/bailian_rerank_source.py
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
---------
Co-authored-by: Ruochen Pan <badbatch0x01@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
* fix(gsvi_tts): Use the correct calling method (#5638)
Add some configuration items for GSVI
* fix(gsvi_tts): add default value for api_key in provider configuration
* fix(gsvi_tts): Adjust wherever the Authorization header is built to only include it when `self.api_key` is truthy
Delete some comments
* chore: ruff format
---------
Co-authored-by: Soulter <905617992@qq.com>
* fix(provider): add missing index field to streaming tool_call deltas
- Fix#6661: Streaming tool_call arguments lost when OpenAI-compatible proxy omits index field
- Gemini and some proxies (e.g. Continue) don't include index field in tool_call deltas
- Add default index=0 when missing to prevent ChatCompletionStreamState.handle_chunk() from rejecting chunks
Fixes#6661
* fix(provider): use enumerate for multi-tool-call index assignment
- Use enumerate() to assign correct index based on list position
- Iterate over all choices (not just the first) for completeness
- Addresses review feedback from sourcery-ai and gemini-code-assist
---------
Co-authored-by: Yaohua-Leo <3067173925@qq.com>
Co-authored-by: Soulter <905617992@qq.com>
* fix: auto-append /v1 to embedding_api_base in OpenAI embedding provider (#6855)
When users configure `embedding_api_base` without the `/v1` suffix,
the OpenAI SDK does not auto-complete it, causing request path errors.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: ensure API base URL for OpenAI embedding ends with /v1 or /v4
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Soulter <905617992@qq.com>
* fix: 解决 MCP 工具与内置工具重名时的连坐问题
- 修改 get_func 方法：优先返回已激活的工具
- 修改 get_full_tool_set 方法：使用 add_tool 防止同名冲突
- 修改 add_tool 方法：优先保留已激活的工具
Fixes#5821
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: address PR review feedback for tool conflict resolution
- Fix inconsistency: get_func now uses reversed() to match ToolSet.add_tool's
"last-active-wins" logic, preventing potential "tool hijacking" issues
- Improve readability: replace double negative condition with clearer logic
- Add compatibility: use getattr with default for tools without 'active' attribute
- Remove unnecessary deepcopy: MCPTool runtime objects should not be deep copied
- Update docstring: accurately describe the actual tool resolution behavior
Addresses review comments from sourcery-ai, gemini-code-assist, and Copilot.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* test: add tests for tool conflict resolution (issue #5821)
Add comprehensive tests for ToolSet.add_tool, get_func, and get_full_tool_set
to verify the conflict resolution behavior when MCP tools share names with
built-in tools.
Test cases:
- ToolSet.add_tool: active/inactive priority, last-one-wins for same state
- get_func: returns last active tool, fallback to last matching tool
- get_full_tool_set: deduplication logic, no deepcopy, MCP overrides disabled builtin
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: 修复工具冲突处理逻辑，确保未激活工具不被错误移除
---------
Co-authored-by: whatevertogo <whatevertogo@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Add Kimi Code provider
* Add icon mapping for Kimi Code provider
* Clarify Kimi CodingPlan provider labeling
* Refine Kimi Code header handling
* modified docker compose
* fix: correct Kimi Coding Plan label and update API base URL
---------
Co-authored-by: Soulter <905617992@qq.com>
* fix: only pass dimensions param when explicitly configured
Models like bge-m3 don't support the dimensions parameter in the
embedding API, causing HTTP 400 errors. Previously dimensions was
always sent with a default value of 1024, even when the user never
configured it. Now dimensions is only included in the request when
embedding_dimensions is explicitly set in provider config.
Closes#6421
Signed-off-by: JiangNan <1394485448@qq.com>
* fix: handle invalid dimensions config and align get_dim return
- Add try-except around int() conversion in _embedding_kwargs to
gracefully handle invalid embedding_dimensions config values
- Update get_dim() to return 0 when embedding_dimensions is not
explicitly configured, so callers know dimensions weren't specified
and can handle it accordingly
- Both methods now share consistent logic for reading the config
Signed-off-by: JiangNan <1394485448@qq.com>
* fix: improve logging for invalid embedding_dimensions configuration
---------
Signed-off-by: JiangNan <1394485448@qq.com>
Co-authored-by: Soulter <905617992@qq.com>
Fixes#6283
When adding a new embedding provider, the knowledge base creation page
did not show the new provider until restart.
Root cause: create_provider() did not update self.providers_config,
which is used by get_provider_config_list() to return provider lists.
This fix syncs the in-memory config after loading the new provider,
consistent with how reload() handles config updates.
Co-authored-by: ccsang <ccsang@users.noreply.github.com>
* Fix CreateSkillPayloadTool array schema missing items field
The payload parameter's anyOf array variant lacked an items field,
causing Gemini API to reject the tool declaration with 400 Bad Request:
'parameters.properties[payload].any_of[1].items: missing field.'
Add items: {type: object} to the array variant to satisfy the Gemini
API requirement for array type schemas.
Fixes#6279
* Fix TypeError when OpenAI-compatible API returns null choices
Some providers (e.g. OpenRouter) may return a completion where
choices is None rather than an empty list — for instance on rate
limiting, content filtering, or transient errors. The existing code
used len(completion.choices) which throws TypeError on None.
Replace all len(...choices) == 0 checks with 'not ... .choices' which
handles both None and empty list. Affects _query_stream, _parse_openai_completion,
and _extract_reasoning_content.
Fixes#6252
In _handle_api_error(), when a 429 rate-limit is encountered, the code
calls available_api_keys.remove(chosen_key). If the same key was already
removed in a previous retry iteration (e.g. the key rotated back to the
same value), this raises ValueError which crashes the entire LLM request
with an opaque error instead of a proper retry/fallback.
Add a membership check before calling remove() to prevent the crash.
Co-authored-by: easonysliu <easonysliu@tencent.com>
- Add audio data validation in MiniMax TTS get_audio() method to detect empty responses
- Validate generated audio file size in TTSProvider.test() to ensure valid output
- Provide detailed error messages guiding users to check group_id configuration
- Auto-cleanup test audio files after validation
- Fixes issue where 0KB audio files would pass TTS detection when group_id is not configured
