- Add ExaWebSearchTool (web_search_exa) with keyword/semantic search,
category filters, domain restrictions, and date range support
- Add ExaGetContentsTool (exa_get_contents) for extracting web page content
- Add _exa_search() and _exa_get_contents() API helpers hitting
https://api.exa.ai/search and https://api.exa.ai/contents
- Add _EXA_KEY_ROTATOR for multi-key rotation
- Register Exa tools in _apply_web_search_tools() dispatch
- Add Exa to WEB_SEARCH_CITATION_TOOL_NAMES for citation support
- Add websearch_exa_key config default and provider option
- Add i18n metadata for en-US, zh-CN, ru-RU
- Add Exa section to docs (en + zh)
- Add 6 unit tests covering search, contents, error handling, and
legacy config migration
Closes#5621
* fix: unify media reference handling
* fix: accept bare base64 record media refs
* chore: update agents.md
* fix: unify file URI handling across media components and utilities
* fix: unify media reference type handling with MediaRefStr alias
* Potential fix for pull request finding 'CodeQL / Incomplete URL substring sanitization'
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
* Update astrbot/core/platform/sources/discord/discord_platform_adapter.py
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
* fix: unify media handling and improve base64 decoding across components
* fix: simplify client_kwargs type definition and enhance media message handling in platform adapter
* fix: unify media utility documentation and enhance function descriptions
* perf: drop "pilk" requirement, improve audio outbound for tencent-related IM apps which using silk
* fix: unify Tencent Silk audio handling and enhance media resolver functionality
---
- Centralize media reference materialization and base64 resolution for local paths, http(s), base64://, data URIs, and legacy bare base64 payloads.
- Normalize incoming Record audio to wav and Image media to temporary jpg during preprocess, with event-scoped cleanup.
- Reuse the shared media resolver across OpenAI, Gemini, Anthropic, MiMo, DeerFlow, STT, and platform media paths while sanitizing logs and cleaning temporary conversion outputs.
- Ensure generated TTS audio is tracked for cleanup after the event finishes.
fix#8676fix#8543fix#7588fix#7580fix#8030fix#8034fix#7461fix#7565fix#6509fix#7144fix#7795
---------
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
* fix(context): restore turn cap, serialize content parts and tool calls for llm compress, fix AftCompact debug log
Three context-compaction regression fixes after #8226:
1. Restore max_context_length -> enforce_max_turns propagation so
normal turn-based truncation works again.
2. Serialize ContentPart and ToolCall objects into plain dicts in
_message_to_dict so llm_compress no longer fails with JSON
serialization errors.
3. Print _provider_messages (compacted) instead of run_context.messages
(unchanged) in AftCompact debug log; truncate long role lists to
first4,...,last4 to avoid log spam.
Assertions in tests are also hardened to avoid coupling to exact prompt
wording.
* fix(tool_loop_agent_runner): simplify context handling by removing redundant provider messages
* fix(tool_loop_agent_runner): rename context manager variables for clarity
* fix: update context compression to use recent token ratio instead of fixed count
* fix: enhance LLMSummaryCompressor to sanitize contexts and improve message handling
* ruff format
---------
Co-authored-by: Soulter <905617992@qq.com>
* refactor(ltm): redesign long-term memory with context compaction
- Add raw_records / contexts / summaries data model per group
- Add LLM summary compaction strategy alongside truncation
- Add turn-based (_split_into_rounds) granularity
- Add image caption integration into LTM history
- Add tool_call / tool_result persistence into raw_records
- Add active reply support driven by LTM state
- Improve summary injection prefix with system note and delimiters
- Add info-level logging for summary compaction lifecycle
- Clarify default summary prompt with explicit preserve/drop rules
- Add context_guard for history overflow protection in agent runner
- Add internal agent history compaction in agent_sub_stages
- Add comprehensive LTM unit tests and compaction test suites
* fix(ltm): handle malformed JSON in tool args and clean up lock on session removal
* fix(ltm): guard against duplicate system prompt note injection
* fix(ltm): fall back to user message when internal marker parsing fails
- Treat lines starting with <T:CALL>, <T:RES, or <BOT/ as regular user
messages when their respective parsers return None, instead of silently
dropping them. Defensive guard against malformed internal markers.
* fix(ltm): release session lock during LLM summary generation
* fix(ltm): trim raw_records in handle_message to prevent unbounded growth
* perf(ltm): use len(s) instead of len(s.encode()) in trim loop
Avoid allocating a new bytes object for every string when calculating
buffer size in _trim_raw_records. Character count is sufficient for
the approximate memory cap.
* feat(ltm): make user segment truncation limits configurable
* feat(ltm): pre-fill default LTM summary prompt in config and i18n
* refactor(ltm): hardcode internal segment/trim constants
* refactor(ltm): unify compaction strategy with main agent runner
* feat(ltm): add @mention weight marker for group chat messages
* test: fix test failures from LTM compaction unification
* chore(dashboard): remove obsolete LTM compaction i18n metadata
* chore: shrink codebase
* feat(group-chat): implement group chat context management and related functionality
---------
Co-authored-by: Tsukumi <112180165+Tsukumi233@users.noreply.github.com>
Co-authored-by: zenfun <zenfun510@gmail.com>
Co-authored-by: Soulter <905617992@qq.com>
* feat: supports plugin to add skills
* fix tests
* fix: fs tools
* Add tests for plugin skills handling and improve skill management
- Implement test for restricted local member reading plugin skill inventory even if the plugin is inactive.
- Ensure that the skill synchronization process retains built-in skills when local skills are empty, including proper handling of plugin paths.
- Update dashboard tests to verify that plugin details include components when requested.
- Enhance skill metadata enrichment tests to include inactive plugin-provided skills for inventory.
- Add filtering tests for plugin skills based on current configuration, ensuring only allowed plugins are considered and inactive plugins are skipped.
Co-authored-by: Copilot <copilot@github.com>
* fix: handle PPIO platform context-length error messages (#7888)
* fix: 压缩算法删除 user 消息 Bug 修复
* perf: improve truncate algo
* fix: improve context length error detection for PPIO platform compatibility
- Extend error detection to handle PPIO's error message format:
'The input is longer than the model's context length'
- Add case-insensitive matching using .lower() for robustness
- Maintain backward compatibility with existing 'maximum context length' check
This fixes the issue where PPIO platform models (e.g., ppio/zai-org/glm-5-turbo)
would fail with AgentState.ERROR due to unrecognized context length errors.
---------
Co-authored-by: Soulter <905617992@qq.com>
* fix: 支持微信客服文件消息 (#7923)
* fix: 支持微信客服文件消息
* fix: remove WeCom file message placeholder
* fix(provider): fix Anthropic custom headers and system prompt compatibility (#7587)
* fix(provider): fix Anthropic custom headers and system prompt compatibility
- Pass custom_headers via AsyncAnthropic's `default_headers` parameter
instead of creating a separate httpx.AsyncClient. This avoids
`isinstance` check failures when multiple httpx installations exist
on sys.path (e.g. bundled Python + system Python).
- Use list format for the `system` parameter (`[{"type": "text", ...}]`)
instead of a plain string. The list format is supported by the official
Anthropic API and is also compatible with third-party API proxies that
reject the string format.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(provider): fix Anthropic custom headers and system prompt compatibility
- Pass custom_headers via AsyncAnthropic's `default_headers` parameter
instead of creating a separate httpx.AsyncClient. This avoids
`isinstance` check failures when multiple httpx installations exist
on sys.path (e.g. bundled Python + system Python).
- Use list format for the `system` parameter (`[{"type": "text", ...}]`)
instead of a plain string. The list format is supported by the official
Anthropic API and is also compatible with third-party API proxies that
reject the string format.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Add test unit
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* perf: improve logic of adding models
Co-authored-by: piexian <piexian@users.noreply.github.com>
* chore: remove redundant logger messages and improve log clarity
Co-authored-by: Copilot <copilot@github.com>
* chore: ruff format
* docs: update knowledge base docs
closes: #7962
* fix(#7907): send_message_to_user cron 场景下 session 容错 (#7911)
* fix: send_message_to_user cron 场景下 session 容错 (#7907)
- LLM 在主动场景可能只传 session_id 而非完整三段式，
from_str 失败时用 current_session 补全前两段。
Co-authored-by: Copilot <copilot@github.com>
* fix: 限制 session 补全仅对裸 session_id 生效，避免误修带冒号的错误输入 (#7907)
* feat: add session information to cron job payload
Co-authored-by: Copilot <copilot@github.com>
* fix: improve clarity and consistency of safety mode prompts
Co-authored-by: Copilot <copilot@github.com>
---------
Co-authored-by: Copilot <copilot@github.com>
Co-authored-by: Weilong Liao <37870767+Soulter@users.noreply.github.com>
Co-authored-by: Soulter <905617992@qq.com>
* perf: tool rendering in conversation page (#7937)
* fix(dashboard): route conversation history tool messages through ToolCallCard
When viewing conversation history, large tool outputs (e.g. a single
git log --stat producing tens of KB) caused the browser renderer to
freeze. Root cause: formattedMessages mapped every role (including
tool / system / _checkpoint) into user/bot bubbles, and bot plain
strings went through markstream-vue's MarkdownRender. Single 88KB
tool messages plus 88-of-them adding up to ~349KB of synchronous
markdown parsing was enough to block the main thread for 5+ seconds.
This patch:
- Indexes tool-role messages by tool_call_id
- Filters formattedMessages to user/assistant only — tool, system and
_checkpoint roles no longer render as standalone bubbles
- Converts assistant.tool_calls (OpenAI shape, with tc.name/tc.arguments
fallbacks) into the existing tool_call MessagePart, attaching the
paired result so MessageList's ToolCallCard renders it (default
collapsed, no longer feeds large strings into the markdown renderer)
- Drops empty placeholder plain parts when an assistant message only
carries tool_calls
- Sets ts/finished_ts to 0 as a sentinel: ToolCallCard.toolCallDuration
returns "" when startTime <= 0, suppressing a misleading "0ms"
duration that would otherwise appear because conversation history
has no real timing data
Behavior change: tool results are now embedded in their assistant's
ToolCallCard.result instead of appearing as separate bot bubbles.
This matches the main chat UI's behavior.
Fixes#7929
Refs #7372#7456
* style(dashboard): use single scrollbar in conversation history preview
ToolCallCard's result/args panes have their own max-height + overflow,
which produced a nested scrollbar when nested inside the history
preview's already-scrollable .conversation-messages-container. Override
those constraints inside the preview only — the outer 500px-bounded
container already provides scroll bounds, so a single scrollbar feels
cleaner. The main chat UI is unaffected.
---------
Co-authored-by: wanger <wanger@example.com>
* fix: ruff format
* feat: add python tool timeout param (#7953)
* feat: add python tool timeout param
* Update python.py
---------
Co-authored-by: Weilong Liao <37870767+Soulter@users.noreply.github.com>
* fix: 钉钉连接超时后自动重连失败 (#7924)
* fix: improve DingTalk adapter error handling in run() method
* fix: add retry logic for DingTalk SDK task unexpected exit
* fix: use task.add_done_callback to wake thread on task completion, handle UnboundLocalError
* refactor: extract retry logic into handle_retry helper function
---------
Co-authored-by: Blueteemo <Blueteemo@users.noreply.github.com>
---------
Co-authored-by: Copilot <copilot@github.com>
Co-authored-by: leonforcode <leonbeyourside01@gmail.com>
Co-authored-by: AstralSolipsism <134063164+AstralSolipsism@users.noreply.github.com>
Co-authored-by: Pink YuDeer <wer00001@outlook.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: piexian <piexian@users.noreply.github.com>
Co-authored-by: NayukiMeko <ChibaNayuki@163.com>
Co-authored-by: wanger <122891289+10knamesmore@users.noreply.github.com>
Co-authored-by: wanger <wanger@example.com>
Co-authored-by: Haoran Xu <3230105281@zju.edu.cn>
Co-authored-by: 千岚之夏 <108566281+Blueteemo@users.noreply.github.com>
Co-authored-by: Blueteemo <Blueteemo@users.noreply.github.com>
- New config item fallback_max_context_tokens (default 128k)
- When max_context_tokens is 0 and model not in LLM_METADATAS,
use fallback_max_context_tokens as the context window limit
- Unified global config under provider_settings, in truncate_and_compress section
- i18n: zh-CN, en-US, ru-RU
Co-authored-by: AstrBot <astrbot@container>
* fix: guard against None system_prompt in _ensure_persona_and_skills
ProviderRequest.system_prompt defaults to None. When a persona with a
prompt is configured, _ensure_persona_and_skills calls
``req.system_prompt += ...`` which crashes with ``TypeError`` when
system_prompt is None.
Added a None guard before the persona prompt injection and skills prompt
appending sections so they always operate on a string.
* chore: delete tests/unit/test_system_prompt_none_bug.py
---------
Co-authored-by: Weilong Liao <37870767+Soulter@users.noreply.github.com>
* feat: add Firecrawl web search and extract tools, update configuration and tests
* feat: implement Firecrawl API integration and error handling in web search tools
* feat: enhance Firecrawl web search with session management and payload validation
* feat: Firecrawl web search to use aiohttp.ClientSession directly for improved session management as it was
* feat: update Firecrawl search to handle grouped web data response and add corresponding tests
* feat: refactor Firecrawl web search to use aiohttp.ClientSession for improved error handling and session management
* feat: remove unused coercion function and update Firecrawl search to use default limit in payload
* feat: add inline message editing and regeneration functionality for webui
- Implemented inline editing for user messages in the chat component.
- Added a regenerate menu for retrying messages with different models.
- Enhanced message handling to include llm_checkpoint_id for better tracking.
- Updated localization files to include new actions for retrying and model selection.
- Introduced tests for checkpoint message handling and chat route functionality.
* feat: thread mode in webui
* feat: enhance message editing functionality to allow only the latest user message to be edited
* feat: add error handling and user feedback for thread creation in chat component
* feat: add thread count display and localization support in chat component
* feat: add RefsSidebar component and integrate reference management in chat UI
* feat: improve message editing validation and cleanup for bot messages
* feat: enhance checkpoint message handling with binding and dumping functionality
* feat: update FileReadTool description to mention image and PDF support
Add explicit mention of image (OCR) and PDF (text extraction) support
to the FileReadTool description for better discoverability.
* feat: update FileReadTool description to include support for docx and epub files; change base64 decoding to utf-8
* feat: enhance ToolLoopAgentRunner to support image and audio modalities; add context sanitization logic
* feat: filesystem grep, read, edit file
* feat: add file write tool and enhance file read functionality
* feat: enhance tool prompt formatting and add escaped text decoding for file editing
* feat: remove redundant safe path tests from security restrictions
* feat: implement file read tool with support for text and image files, including validation for large files
* feat: add file read utilities and integrate with filesystem tools
* refactor: move computer tools to builtin tools registry
* refactor: remove unused plugin_context parameter from _apply_sandbox_tools
* feat: supports to display enabled builtin tools in configs
* feat: add tooltip for disabled builtin tools and update localization strings
* feat: add workspace extra prompt handling in message processing
* feat: add ripgrep installation to Dockerfile
* perf: shell executed in workspace dir in local env
* feat: enhance file reading capabilities to support PDF and DOCX parsing, including workspace storage for long documents
* feat: update converted text notice to suggest using grep for large files
* feat: implement handling for large tool results with overflow file writing and read tool integration
* fix: test
* feat: enhance onboarding steps to include computer access configuration and related help information
* feat: add support for additional temporary path in restricted environment checks
* feat: update computer access hints and add detailed configuration instructions
* perf: replace cron tools with FutureTaskTool for improved task management
* feat: enhance FutureTaskTool with edit functionality and improve descriptions
* feat: add edit functionality for cron jobs and update related UI components
* 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
* 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]"
* fix: resolve subagent persona lookup for 'default' and unify resolution logic
- Add PersonaManager.get_persona_v3_by_id() to centralize v3 persona resolution
- Handle 'default' persona_id mapping to DEFAULT_PERSONALITY in subagent orchestrator
- Fix HandoffTool.default_description using agent_name parameter correctly
- Add tests for default persona in subagent config and tool deduplication
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: simplify get_default_persona_v3 using get_persona_v3_by_id
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: whatevertogo <whatevertogo@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Soulter <37870767+Soulter@users.noreply.github.com>
Add explicit Neo lifecycle instructions to the main agent prompt so
skill creation and updates follow payload -> candidate -> promotion
instead of direct local folder writes.
Clarify lifecycle tool descriptions and parameter semantics, including
skill_key/source_execution_ids usage and stable release sync_to_local
behavior, to reduce ambiguity and improve consistent skill publishing.
append a Shipyard Neo-specific system prompt note for filesystem
tool calls so paths are provided relative to the workspace root.
this prevents models from prepending `/workspace` and causing tool
path resolution failures
_apply_sandbox_tools now checks the booted session's capabilities
before registering browser tools (BrowserExecTool, BrowserBatchExecTool,
RunBrowserSkillTool).
- If no session exists yet (first request), all tools are registered
conservatively to avoid breaking the initial interaction
- If a session exists without browser capability, browser tools are
omitted, preventing CapabilityNotSupportedError from Bay
- Skill lifecycle tools remain unconditionally registered
* feat: implement fallback provider support for chat models and update configuration
* feat: enhance provider selection display with count and chips for selected providers
* feat: update fallback chat providers to use provider settings and add warning for non-list fallback models
