v4.23.0-beta.1

• 更新日志(简体中文)
• Changelog(English)

What's Changed

新增

• 为电脑使用能力支持文件读取(read)、写入(write)、编辑(edit)、Grep 搜索(ripgrep)与按会话隔离的 workspace。（#7402）
• 新增 Brave Search 网页搜索工具，替代旧的默认网页搜索实现。（#6847）
• 新增 Mattermost 平台适配器支持。（#7369）
• 新增 NVIDIA Rerank Provider。（#7227）
• 新增 OpenAI、Gemini 音频输入模态支持，并修复 ChatUI 录音相关问题。（#7378）
• Discord 平台新增 Bot 消息过滤配置，允许控制是否接收其他 Bot 的消息。（#6505）
• 新增 LLM 对重复工具调用的引导能力，减少模型陷入重复调用工具的情况。（#7388）

优化

• 合并 Cron 相关工具为统一管理工具，并新增 Cron 任务编辑能力。（#7445）
• 重构内置工具管理逻辑，改善 AstrBot 内置工具注册与调用维护性。（#7418）
• 移除默认网页搜索实现，改由新的搜索工具链路提供能力。（#7416）
• 移除 lxml 与 beautifulsoup4 依赖，降低安装体积与依赖复杂度。（#7449）
• 降低 MCP Server 状态轮询频率，减少后台请求开销。（#7399）
• 修正文档中的路径拼接示例，避免插件开发存储文档误导使用者。（#7448）

修复

• 修复 Windows 桌面端插件依赖加载不安全或失败的问题。（#7446）
• 修复 Telegram 长消息最终分段过长的问题。（#7432）
• 修复 Telegram sendMessageDraft 发送空文本导致 400 错误刷屏的问题。（#7398）
• 修复 Telegram 收集命令时插件 handler 不在 star_map 中导致 KeyError 的问题。（#7405）
• 修复 QQ 官方 WebSocket 关闭流程清理不完整的问题。（#7395）
• 修复 Gemini FunctionResponse 中不支持的 id 字段导致请求失败的问题。（#7386）
• 修复 Gemini 空模型输出误触发错误处理的问题。（#7377）
• 修复仅存在原生工具时仍传递 FunctionCallingConfig 的问题。（#7407）
• 修复 ChatUI 项目常量缺失，并补充相关测试用例。（#7414）
• 修复 WebUI 暗色模式渲染与多处交互问题。（#7173）
• 修复页面切换时浮动按钮跳动的问题。（#7214）
• 修复 faiss 在启动阶段过早导入导致部分环境启动失败的问题。（#7400）
• 修复微信个人号适配器缺少上下文 token 时的警告信息。（commit）
• 修复工具结果断言与动态阈值不一致的问题。（commit）

What's Changed (EN)

New Features

• Added local Computer Use filesystem tools, including file read, write, edit, Grep search, and per-session workspace support. (#7402)
• Added the Brave Search web search tool, replacing the old default web search implementation. (#6847)
• Added Mattermost platform support. (#7369)
• Added the NVIDIA Rerank Provider. (#7227)
• Added audio input support across providers and fixed ChatUI recording issues. (#7378)
• Added configurable Discord bot-message filtering, including support for receiving messages from other bots. (#6505)
• Added LLM guidance for repeated tool calls to reduce repetitive tool-call loops. (#7388)

Improvements

• Merged Cron tools into a unified management tool and added Cron task editing. (#7445)
• Refactored built-in tool management to improve registration and maintenance. (#7418)
• Removed the old default web search implementation in favor of the new search tool flow. (#7416)
• Removed lxml and beautifulsoup4 dependencies to reduce dependency weight and installation complexity. (#7449)
• Reduced MCP Server status polling frequency to lower background request overhead. (#7399)
• Fixed a path concatenation example in the plugin storage docs. (#7448)
• Updated the README logo. (commit)

Bug Fixes

• Fixed safer desktop plugin dependency loading on Windows. (#7446)
• Fixed overly long final Telegram message segments. (#7432)
• Fixed Telegram sendMessageDraft 400 spam caused by empty text. (#7398)
• Fixed a KeyError in Telegram command collection when a plugin handler is missing from star_map. (#7405)
• Cleaned up QQ Official WebSocket shutdown handling. (#7395)
• Removed the unsupported id field from Gemini FunctionResponse payloads. (#7386)
• Fixed empty model output handling that could misfire when using Gemini. (#7377)
• Skipped FunctionCallingConfig when only native tools are present. (#7407)
• Added missing ChatUI project constants and related tests. (#7414)
• Fixed WebUI dark-mode rendering and multiple interaction bugs. (#7173)
• Prevented floating buttons from jumping during page transitions. (#7214)
• Deferred faiss imports during startup to avoid startup failures in some environments. (#7400)
• Improved the missing-context-token warning message in the Weixin OC adapter. (commit)
• Updated tool-result assertions to match dynamic threshold values. (commit)

What's Changed

• docs: add usage of select_knowledgebase in plugin development docs by @piexian in #7115
• feat: add new StatsPage for enhanced statistics overview by @Soulter in #7152
• fix(core): Record.path was incorrectly treated as required under Pydantic v2 by @No-22-Github in #7048
• fix: improve number input UX in config renderer by @Reisenbug in #7153
• fix: send SSE heartbeat to prevent WebChat disconnect during compression by @he-yufeng in #7003
• perf: improve ui by @Soulter in #7178
• chore(deps): bump codecov/codecov-action from 5 to 6 in the github-actions group by @dependabot[bot] in #7209
• Fix cached_tokens handling in _extract_usage method by @LIghtJUNction in #6719
• refactor: update provider panels for improved layout and styling by @Soulter in #7248
• feat: integrate Monaco Editor workers for enhanced code editing support by @Soulter in #7249
• fix: add pysocks dependency to support SOCKS5 proxy for pip install by @Neko-Yukari in #7221
• fix(Docker): add amr encoder by @WenqiOfficial in #7247
• fix: support both old and new Bailian Rerank API response formats by @he-yufeng in #7217
• 修正了QQ websoket连接的文档错误 by @Thelia-Lzr in #7279
• 修复：过滤空助手消息，以防止在严格API上出现400错误（fix: filter empty assistant messages to prevent 400 error on strict APIs） by @kaixinyujue in #7202
• Docs/update mimo provider readme by @RichardLiuda in #7207
• Feat(webui): dashboard and console qol improvements by @M1LKT in #7215
• fix: 改进知识库的初始化错误处理 by @piexian in #7243
• docs: 在 uv 部署文档中添加不支持 WebUI 升级的说明 by @piexian in #7298
• fix: resolve Discord/Misskey hot reload issue by fixing client_self_id misuse by @Soulter in #7331
• fix: satisfy Google Gemini's function_response requirements to avoid 400 Invalid argument errors by @he-yufeng in #7216
• fix: filter Gemini thinking parts from user-facing message chain by @he-yufeng in #7196
• perf: Set content to None when the OpenAI message content list is empty by @he-yufeng in #6551
• fix(windows): inherit all system environment variables into MCP runner for Windows by @SoloLevelingAI in #7054
• feat: support token usage extraction for llama.cpp by @Soulter in #7358
• feat(provider/vllm_reran...

v4.22.3

• 更新日志(简体中文)
• Changelog(English)

What's Changed

新增

• 新增 vllm_rerank 提供商支持可配置的 rerank_api_suffix，可灵活适配不同 API 路径。（#7278）
• 适配 llama.cpp 的 token 用量提取支持，增强 token 统计与计费分析能力。（#7358）
• WebUI 捆绑 Monaco Editor 的 worker 文件，防止从 CDN 下载失败。（#7249）
• 全新的模型/平台总览统计面板。（#7152）

优化

• 优化 OpenAI 消息处理，若 content 列表为空时将其置为 None，并降低空内容下的异常率。（#6551）
• 优化 provider page 布局与样式，提升整体设置页面排版一致性。（#7248）
• 用户已登录时将直接跳转欢迎页，减少再次进入登录页的步骤。（commit）

修复

• 修复 Windows 下 MCP runner 未继承完整系统环境变量导致启动异常的问题。（#7054）
• 修复 Gemini 思考片段泄露到用户可见消息链路的问题。（#7196）
• 修复 Gemini(OpenAI 兼容 API) 下，function_response 格式要求导致的 400 Invalid argument 错误。（#7216
• 修复知识库初始化失败时的错误提示与恢复问题。（#7243）
• 修复 Bailian Rerank 对新旧响应格式兼容性，避免解析失败。（#7217）
• Docker 镜像补充 AMR 编码器依赖，修复部分部署场景下的音频转码缺失问题。（#7247）
• 修复 pip 安装缺少 pysocks 依赖导致的 SOCKS5 代理不可用问题。（#7221）
• 修复 cached_tokens 在 _extract_usage 中的处理逻辑，避免 token 统计错乱。（#6719）
• 增强 SSE 断开保护，压缩传输下主动发送心跳，减少 WebChat 误断连。（#7003）
• 修复严格 API 下空 assistant 消息导致的 400 错误。（#7202）
• 新增二维码功能缺失的 qrcode 依赖修复，恢复二维码生成功能。（commit）
• 修复配置器数字输入组件的交互体验问题，避免异常输入场景下的处理偏差。（#7153）
• 修复 Record.path 在 Pydantic v2 下被错误视为必填字段的问题。（#7048）

What's Changed (EN)

New Features

• Added configurable rerank_api_suffix for the vllm_rerank provider to support custom API paths.（#7278）
• Added token usage extraction support for llama.cpp to improve token accounting.（#7358）
• Bundled Monaco Editor workers in WebUI so code editing does not depend on external CDN downloads.（#7249）
• Added a new statistics overview for model and platform usage.（#7152）
• Redirected logged-in users to the welcome page instead of the login page.（commit）

Improvements

• Improved OpenAI message handling by setting message content to None when content lists are empty, reducing empty-payload edge cases.（#6551）
• Improved provider page layout and styling to keep settings pages consistent.（#7248）
• Refined connection-success handling and messaging when users are already logged in.（commit）

Bug Fixes

• Inherited all system environment variables in Windows MCP runner to fix environment-dependent launch issues.（#7054）
• Filtered Gemini thinking parts from user-facing message chains.（#7196）
• Fixed Gemini function_response compatibility to prevent 400 Invalid argument responses.（#7216）
• Improved knowledge base initialization error handling.（#7243）
• Fixed Bailian rerank parsing for both old and new response formats.（#7217）
• Added AMR encoder support in Docker builds to fix missing audio codec support.（#7247）
• Added the pysocks dependency so pip installs can use SOCKS5 proxy correctly.（#7221）
• Fixed _extract_usage cached token handling.（#6719）
• Sent SSE heartbeat packets to prevent WebChat disconnection under compression.（#7003）
• Fixed 400 errors caused by empty assistant messages on strict APIs.（#7202）
• Added missing qrcode dependency to restore QR code generation.（commit）
• Fixed number input UX in config renderer.（#7153）
• Fixed Record.path being incorrectly treated as required under Pydantic v2.（#7048）

What's Changed

• docs: add usage of select_knowledgebase in plugin development docs by @piexian in #7115
• feat: add new StatsPage for enhanced statistics overview by @Soulter in #7152
• fix(core): Record.path was incorrectly treated as required under Pydantic v2 by @No-22-Github in #7048
• fix: improve number input UX in config renderer by @Reisenbug in #7153
• fix: send SSE heartbeat to prevent WebChat disconnect during compression by @he-yufeng in #7003
• perf: improve ui by @Soulter in #7178
• chore(deps): bump codecov/codecov-action from 5 to 6 in the github-actions group by @dependabot[bot] in #7209
• Fix cached_tokens handling in _extract_usage method by @LIghtJUNction in #6719
• refactor: update provider panels for improved layout and styling by @Soulter in #7248
• feat: integrate Monaco Editor workers for enhanced code editing support by @Soulter in #7249
• fix: add pysocks dependency to support SOCKS5 proxy for pip install by @Neko-Yukari in #7221
• fix(Docker): add amr encoder by @WenqiOfficial in #7247
• fix: support both old and new Bailian Rerank API response formats by @he-yufeng in #7217
• 修正了QQ websoket连接的文档错误 by @Thelia-Lzr in #7279
• 修复：过滤空助手消息，以防止在严格API上出现400错误（fix: filter empty assistant messages to prevent 400 error on strict APIs） by @kaixinyujue in #7202
• Docs/update mimo provider readme by @RichardLiuda in #7207
• Feat(webui): dashboard and console qol improvements by @M1LKT in #7215
• fix: 改进知识库的初始化错误处理 by @piexian in #7243
• docs: 在 uv 部署文档中添加不支持 WebUI 升级的说明 by @piexian in #7298
• fix: resolve Discord/Misskey hot reload issue by fixing client_self_id misuse by @Soulter in #7331
• fix: satisfy Google Gemini's function_response requirements to avoid 400 Invalid argument errors by @he-yufeng in #7216
• fix: filter Gemini thinking parts from user-facing message chain by @he-yufeng in #7196
• perf: Set content to None when the OpenAI message content list is empty by @he-yufeng in #6551
• fix(windows): inherit all system environment variables into MCP runner for Windows by @SoloLevelingAI in #7054
• feat: support token usage extraction for llama.cpp by @Soulter in #7358
• feat(provider/vllm_rerank): add configurable rerank_api_suffix option by @richard950825-sys in #7278

New Contributors

• @No-22-Github made their first contribution in #7048
• @Neko-Yukari made their first contribution in #7221
• @WenqiOfficial made their first contribution in #7247
• @Thelia-Lzr made their first contribution in #7279
• @kaixinyujue made their first contribution in #7202
• @SoloLevelingAI made their first contribution in #7054
• @richard950825-sys made their first contribution in #7278

Full Changelog: v4.22.2...v4.22.3

v4.22.2

What's Changed

新增

• 个人微信 新增“对方正在输入...”状态控制能力。（#6977）
• 配置管理页支持从现有配置复制生成新配置，减少重复创建成本。（#6785）

优化

• Chat Provider text_chat 接口新增 tool_choice 参数，并为模型空响应引入统一的 EmptyModelOutputError 与增强重试逻辑，改善 “skills-like” 工具调用模式下的稳定性。（#7101、#7104）
• SSL 证书配置错误时自动回退到非 SSL 模式，而不是程序直接退出。（#7102）

修复

• 修复 个人微信 Bot 无法发送多媒体文件与 CDN 上传地址兼容性问题，适配新的 upload_full_url 返回格式。#7066）
• 修复 个人微信 连接超时 / 中断后直接导致微信适配器实例退出的问题，增强了重试机制。（#7041
• 修复 WebUI Skills 上传对多技能压缩包的处理，允许单个 zip 包内包含多个技能。（#7070）
• 修复 OpenAI 嵌入模型接口处理相关问题。#7026）
• QQ 机器人发送语音或音频文件没有响应 （#7007）
• ChatUI 切换到某个会话，刷新后错误地跳转到最新会话 （#6535）
• 修复百炼 qwen3-rerank 的请求负载兼容性问题，并忽略不受支持的 return_documents 参数。（#6222）
• 修复 QQ 下，发送唤醒前缀后，多次空触发 LLM 的问题。（#6893）
• 修复 GSVI TTS 调用方式与鉴权头构造问题，并补充默认配置值，适配新的 /infer_single API。（#7083）
• 修复缺失 httpx SOCKS 代理依赖的问题，恢复 SOCKS 代理支持。（#7093）
• 修复企业微信客服发送失败时无法回退普通消息接口的问题。（#7012）
• 修复 WebUI 对话管理详情页无法滚动的问题。#6972）
• 修复 Telegram 附件消息 caption 丢失的问题。（#7020）

What's Changed (EN)

New Features

• Added tool_choice to Tool Loop Agent Runner and introduced EmptyModelOutputError with stronger retry handling, improving stability for "skills-like" tool-call workflows.（#7101、#7104）
• Added SSL config resolution and validation for Dashboard, with automatic fallback to non-SSL mode when the certificate setup is invalid.（#7102）
• Added the ability to create a new config by copying an existing one in the config management page.（#6785）
• Updated the QQ Official adapter to support asynchronous message parsing and attachment preparation.（#7007）
• Added typing-state control ("user is typing...") for the Weixin OC adapter.（#6977）

Improvements

• Refactored Chat UI routing and layout so the UI mode is URL-driven and related state is scoped to sessionStorage.（#6535）
• Improved WebUI polish by centering extension-page toast hints and making code blocks more readable in dark mode.（#6043、#7014）

Bug Fixes

• Fixed skill installer handling so a single zip archive can contain multiple skills.（#7070）
• Fixed Bailian qwen3-rerank payload compatibility and ignored unsupported return_documents parameters.（#6222）
• Fixed pipeline wake-up and request flow on empty messages.（#6893）
• Fixed GSVI TTS invocation and authorization-header construction, and added safer default config values.（#7083）
• Fixed missing httpx SOCKS proxy dependency to restore SOCKS proxy support.（#7093）
• Fixed Weixin OC inbound polling retry behavior and CDN upload compatibility with the new upload_full_url response field.（#7041、#7066）
• Fixed WeCom fallback behavior so sending can retry through the regular message API when the KF API returns 40096.（#7012）
• Fixed missing attachment captions in Telegram messages.（#7020）
• Hardened OpenAI attachment recovery and fixed Embedding URL suffix trimming to avoid recovery failures and accidental URL character removal.（#7004、#7026）
• Fixed Dashboard regressions around list-option label mapping, missing icons, and mouse-wheel handling in the preview container.（#6844、#6970、#6972）

What's Changed

• fix: honor computer_use_require_admin in shipyard_neo tools by @1zzxy1 in #6951
• fix(dashboard): include missing vuetify mdi icons by @RC-CHN in #6970
• docs: Fix typo in plugin configuration documentation by @Yinr in #6971
• docs: normalize QQ group listings in READMEs and community docs by @idiotsj in #6976
• fix(provider): restore parameter transparency in core LLM provider adapters by @SXP-Simon in #6934
• feat(dashboard): center extension page toast hints with the global UI by @catDforD in #6043
• docs: update plugin dev link in webui by @xrr2016 in #6978
• Feat(webui): improve code block readability in dark mode（iss#6963) by @M1LKT in #7014
• fix: harden OpenAI attachment recovery by @zouyonghe in #7004
• revert: "fix(provider): restore parameter transparency in core LLM provider adapters" by @Soulter in #7023
• fix(dashboard): update aiocqhttp tutorial links to new docs path by @idiotsj in #7038
• fix(embedding): use removesuffix instead of rstrip to prevent over-trimming URLs by @Izayoi9 in #7026
• fix(telegram): preserve attachment captions by @Clhikari in #7020
• fix(wecom): fallback to message API when kf returns 40096 by @silwings1986 in #7012
• fix(chatui): refactor routing and layout to drive UI mode from URL and scope state to sessionStorage by @itamiokanjiro in #6535
• fix(weixin_oc): add error handling and retry logic for inbound updates polling by @Soulter in #7041
• feat: Add "typing" ("对方正在输入...") state control for weixin_oc plateform by @GowayLee in #6977
• docs: corrent weixin_oc wechat version requirements by @tlw00988 in #7068
• fix(weixin_oc): allow CDN uploads to use upload_full_url when provided by @Astral-Yang in #7066
• fix: bundle httpx SOCKS proxy support by @zouyonghe in #7093
• fix(gsvi_tts): Use the correct calling method by @Rain-0x01-39 in #7083
• fix(pipeline): skip waking on empty messages by @Reisenbug in #6893
• fix: allow multiple skills in a single zip archive by @jmt059 in #7070
• feat: update QQOfficialPlatformAdapter to support async parsing and attachment preparation by @Soulter in #7007
• fix(dashboard): apply labels mapping for list options in config renderer by @Foolllll-J in #6844
• feat: allow copy config from existing configs by @Flartiny in #6785
• fix(provider): fix Bailian rerank payload compatibility for qwen3-rerank by @kawayiYokami in #6222
• fix(webui): keep wheel scrolling inside history dialog by @1zzxy1 in #6972
• feat(agent-runner): add tool_choice parameter to fix empty tool calls response in "skills-like" tool call mode by @Soulter in #7101
• feat(dashboard): add SSL configuration resolution for dashboard by @Soulter in #7102
• feat: implement EmptyModelOutputError for handling empty responses across providers and enhance retry logic in ToolLoopAgentRunner by @Soulter in #7104

New Contributors

• @Yinr made their first contribution in #6971
• @xrr2016 made their first contribution in #6978
• @silwings1986 made their first contribution in #7012
• @tlw00988 made their first contribution in #7068
• @Astral-Yang made their first contribution in #7066
• @Rain-0x01-39 made their first contribution in #7083
• @jmt059 made their first...

v4.22.1

What's Changed

新增

• 增强 Skills 安装流程，不再限制上传的压缩包顶级必须是一个目录。并支持中文技能名称显示。（#6952）
• OpenAI Embedding 模型配置支持自动补齐 /v1 基础路径。（#6863）
• 为 /api/file 新增 GET 端点并支持多种请求方式。（#6874）
• WebUI 设置页新增日志与缓存清理能力。（#6822）
• Lark 平台新增可折叠 Thinking 面板能力与消息处理优化。（#6831）

修复

• 修复 QQ 官方机器人中，在 Cron Job 或者主动发送消息时的 msg_id 相关负载处理问题。（#6604）
• 修复 个人微信 在轮询超时后停止轮询的问题。（#6915）
• 修复 硅基流动 提供商无法正确使用工具调用能力的问题。（#6829）
• 修复部分提供商工具调用流式增量返回缺少 index 导致的异常。（#6661
• 修复 WebUI 中 ObjectEditor 的 updateKey 错误索引导致的“键已存在”误判。（#6825）
• 修复 UI 图标集合及测试一致性导致的展示异常。（#6894、#6892）
• 修复 T2I 配置间未同步生效模板的问题。（#6824）
• 修复 MIMO TTS 样式参数以对齐官方文档约定。（#6814）

What's Changed (EN)

New Features

• Enhanced skill installation to support multiple top-level folders, duplicate handling, and Chinese skill names.（#6952）
• Automatically append /v1 to embedding_api_base for OpenAI embedding compatibility.（#6863）
• Added plugin author display and pinned plugin card support in WebUI.（#6875）
• Added GET endpoint for /api/file and support for multiple HTTP methods.（#6874）
• Added log and cache cleanup in Dashboard settings.（#6822）
• Added collapsible reasoning panel and message handling improvements for Lark.（#6831）

Improvements

• Validate config_path before existence checks to avoid false negatives.（#6722）
• Improved Provider batch reset behavior and "follow" configuration handling in WebUI.（#6825）
• Updated OpenAI-related guidance in i18n docs for clearer compatibility hints.（adc252a3)

Bug Fixes

• Fixed missing index field in streaming tool_call deltas.（#6661）
• Fixed msg_id payload handling for QQ API.（#6604）
• Kept Weixin OC polling active after inbound timeout.（#6915）
• Fixed updateKey index bug in WebUI ObjectEditor that caused false “key exists” errors.（#6825）
• Fixed icon regressions in UI and related icon scan tests.（#6894、#6892）
• Fixed SiliconFlow provider tools compatibility issue.（#6829）
• Synchronized active T2I template across all configs.（#6824）
• Aligned MIMO TTS payload style with official docs.（#6814）
• Removed privacy-sensitive data left in tests.（#6803）

v4.22.0

What's Changed

新增

• 新增个人微信接入能力（微信官方提供）。详见 个人微信接入（#6777）
• 新增图片压缩能力，支持图片尺寸、JPEG 质量限制。（#6794）
• 新增 WebUI 已安装插件置顶与拖拽重排能力。（#6776）

优化

• 知识库为空时自动跳过检索，避免无效搜索调用。（#6750）
• 修复 frontmatter 转义换行符保留行为并更新相关测试与 CI 工作流（含文档与配置同步）。（#6783）

修复

• Telegram 适配器改为导入 Forbidden，替代已弃用的 Unauthorized（兼容性修正）。（#6769）
• 处理 OpenAI Provider 中 token usage metrics 为 None 时的异常场景。（#6788）
• 保留全部 CallToolResult 的 content 项，避免工具调用上下文内容丢失。（#6149）

What's Changed (EN)

New Features

• Added WeChat personal account support, including QR-code driven setup, message flow refactor, and media handling improvements.（#6777）
• Added pinning and drag-sorting for installed WebUI plugins.（#6776）
• Added image compression with max-size handling and temporary-file lifecycle management for improved image efficiency.（#6794）

Improvements

• Skipped search execution when the entire knowledge base is empty.（#6750）
• Preserved escaped newlines in frontmatter, with related test and CI workflow updates.（#6783）

Bug Fixes

• Replaced deprecated Telegram adapter import of Unauthorized with Forbidden.（#6769）
• Handled None values in OpenAI provider token usage metrics.（#6788）
• Kept all CallToolResult.content items to prevent losing tool-call context.（#6149）

What's Changed

• feat: supports weixin personal account by @Soulter in #6777
• fix(platform.tg_adapter): import Forbidden instead of deprecated Unauthorized (#6765) by @mzwing in #6769
• feat: skip search when the entire knowledge base is empty by @Waterwzy in #6750
• feat: fix preserve escaped newlines in frontmatter & update tests & ci workflows by @RC-CHN in #6783
• Feat(webui): support pinning and dragging for installed plugins (#6649) by @M1LKT in #6776
• fix: handle potential None values for token usage metrics in OpenAI provider by @Soulter in #6788
• feat: supports image compressing (#6463) by @Soulter in #6794
• fix: keep all CallToolResult content items by @tsubasakong in #6149

New Contributors

• @mzwing made their first contribution in #6769
• @M1LKT made their first contribution in #6776

Full Changelog: v4.21.0...v4.22.0

v4.21.0

What's Changed

新增

• 为插件市场新增分类筛选并优化界面组件。（#6762）
• 新增 Kimi Coding Plan Provider（兼容 Anthropic API）。（#6559）
• 统一并重命名 skill.md 文件为 SKILL.md。（#6757）
• 新增 Ollama 关闭思考模式开关。（#5941）
• 新增小米 MiMo TTS 与 STT Provider。（#6643）
• 新增多模态上下文 token 计数能力（图片、音频与思维链）。（#6596）
• 新增按元信息名称安装插件并校验可导入标识符。（#6530）

优化

• 降低 kook_type 中 StrEnum 到 (str, Enum) 以兼容旧版本 Python。（#6512）
• 优化 WebChat 消息处理：提前保存消息并增强流式补全。（#6698）
• 优化 Dashboard 中移动端 persona 选择器布局。（#5907）
• 优化 Dashboard，减少 MDI icon font 范围并自托管 Google Fonts。（#6532）
• 使用 Pydantic 重构 KOOK 适配器数据模型以提升消息结构解析与校验性能。（#5719）
• 调整文档术语与表达（“消息平台”改为“机器人”等）并优化更新说明文字。（#6568）

修复

• 修复 web search tools 中被禁用系统函数时未正确生效的问题。（#6584）
• 跳过空的知识库嵌入批次，避免不必要计算。（#6106）
• 修复 Feishu opus 文件在 Whisper API STT 场景下未正确转换的问题。（#6078）
• 修复 Lark 在工具调用中断时卡片创建与重续流程。（#6743）
• 修复 Windows 环境下 skill 文件编码兼容问题。（#6058）
• 修复 WebChat 单独渲染的 HTML 回复被错误显示的问题。（#6074）
• 修复 Gemini 数组 schema 缺少 items 字段的问题。（#6051）
• 修复 skills 使用硬编码工作区路径的问题，改为使用缓存中的真实沙箱路径。（#6331）
• 修复 subagent 工具调用超时传递，修正 tool_call_timeout 并将默认超时从 60 秒提升到 120 秒。（#6713）
• 修复 skills-like 下缺失 extra_user_content_parts 时未注入图片说明的问题。（#6710）
• 修复 /stop 后继续处理后续 follow-up 消息的问题。（#6704）
• 修复核心流程在 stop 时中断子代理工具等待的行为。（#5850）
• 修复 Windows 平台 stdio MCP 服务 PATHEXT 丢失的问题。（#5822）
• 修复 Telegram 轮询重启机制，失败时自动恢复循环。（#6648）
• 修复 /stop 后 follow-up 状态未清理导致后续触发继续运行的问题。（#6656）
• 修复 OpenAI 配置日志，改进代理与 API Base 的可读性与排障能力。（#6669）
• 修复长工具调用中截断逻辑误删唯一 user 消息的问题。（#6198）
• 修复截断器移除唯一用户消息导致部分 provider 返回 400 的问题。（#6581）
• 修复 ID 白名单为空时提示文案不清晰的问题。（#6611）
• 修复 MoonshotAI 官方 API 下 OpenAI token usage 缺失问题。（#6618）
• 提升 Wecom AI Bot 流式可读性与速度（限流处理），改善卡顿与显示体验。（#6610）
• 修复 Wecom 长连接回复消失的问题。（#6606）
• 修复 Windows 打包运行时 pip/native 构建环境配置问题。（#6575）
• 修复仅在显式配置时传递 embedding dimensions，避免无效参数干扰。（#6432）
• 修复 register_agent 装饰器导致的 NameError。（#5765）
• 修复默认 persona 场景下的 subagent 查询失败问题。（#5672）
• 修复 Windows 下 skill 读取失败问题。（#6490）
• 修复 SKILL.md 多行 frontmatter 解析问题。（#6460）
• 修复 SQLite database is locked 问题，新增繁忙超时处理。（#6474）
• 修复工作流限制仅允许上游仓库触发的报错场景。（#6531）

---

What's Changed (EN)

New Features

• Added category filtering for market plugins and enhanced UI components.
• Normalized and renamed skill.md files to SKILL.md.
• Added a toggle to disable thinking mode in Ollama.
• Added Xiaomi MiMo TTS & STT providers.
• Added Kimi Coding Plan provider with Anthropic API compatibility.
• Added context token counting support for multimodal content (images, audio, and chain-of-thought).
• Added plugin installation by metadata name and validated importable identifiers.

Improvements

• Refactored StrEnum to (str, Enum) in kook_type for backward compatibility.
• Enhanced WebChat message handling with proactive saving and streaming completion.
• Simplified dashboard persona selector layout for mobile screens.
• Subset MDI icon font and self-hosted Google Fonts in dashboard.
• Implemented Pydantic data models for KOOK adapter for better retrieval and schema validation.
• Updated documentation wording for incomplete updates terminology.

Bug Fixes

• Fixed disabled system functions in web search tools.
• Skipped empty knowledge-base embedding batches.
• Converted Feishu opus files for Whisper API STT.
• Deferred Lark card creation and renewal on tool call break.
• Added fallback for Windows skill file encodings.
• Rendered standalone HTML replies as code in WebChat.
• Ensured Gemini array schemas always include items.
• Fixed skills path resolution to use cached sandbox path instead of hardcoded workspace root.
• Passed tool_call_timeout to subagent handsoff/cron/background tasks and increased default timeout to 120 seconds.
• Fixed skills-like re-query missing extra_user_content_parts causing image caption not injected.
• Rejected follow-up messages after /stop request.
• Interrupted subagent tool waits on stop.
• Preserved PATHEXT for stdio MCP servers on Windows.
• Added automatic restart for Telegram polling loop on failure.
• Fixed follow-up persistence logic after /stop trigger.
• Improved OpenAI proxy/API base logging.
• Prevented truncation logic from removing the only user message in long tool-calling conversations.
• Fixed truncation removing the only user message and returning 400 for some providers.
• Updated ID whitelist empty-configuration hint wording.
• Fixed token usage with MoonshotAI official API.
• Improved Wecom AI Bot streaming readability and speed via throttling.
• Prevented Wecom AI Bot long connection replies from disappearing.
• Set packaged Windows runtime build env for pip native builds.
• Only pass embedding dimensions when explicitly configured.
• Fixed register_agent decorator NameError.
• Fixed subagent lookup failure when using default persona.
• Fixed reading skills on Windows.
• Fixed multiline frontmatter parsing in SKILL.md.
• Fixed SQLite database is locked with busy timeout.
• Restricted workflows to upstream repo.

What's Changed

• feat: localize session management group & interval method texts by @RC-CHN in #6471
• fix: SQLite 'database is locked' under concurrent writes by @he-yufeng in #6474
• fix: parse multiline frontmatter description in SKILL.md by @RhoninSeiei in #6460
• chore(deps): bump the github-actions group with 2 updates by @dependabot[bot] in #6461
• chore: remove deprecated version field from compose.yml by @YuanyuanMa03 in #5495
• fix: reading skills on Windows by @linzhengtian in #6490
• fix: subagent create failure when using default persona by @whatevertogo in #5672
• fix: register_agent decorator NameError by @whatevertogo in #5765
• fix: only pass dimensions when explicitly configured in embedding config by @jnMetaCode in #6432
• perf: Implement Pydantic data models for the KOOK adapter to enhance data retrieval and message schema validation by @shuiping233 in #5719
• refactor: Downgrade StrEnum to (str, Enum) in kook_type for backward compatibility by @shuiping233 in #6512
• feat: install plugin using metadata name and validate importable identifiers by @Soulter in #6530
• fix: restrict workflows to upstream repo by @Raven95676 in #6531
• Fix LLM always choosing FileDownloadTool instead of FileUploadTool by @he-yufeng in #6527
• perf(dashboard): subset MDI icon font and self-host Google Fonts by @camera-2018 in #6532
• 更改未完成更新的文档用词问题（多处“消息平台”已更名为“机器人”） by @nuomicici in #6568
• fix: set packaged Windows runtime build env for pip native builds by @zouyonghe in #6575
• feat (doc) : Add doc for shipyard-neo sandbox driver by @RC-CHN in #6590
• docs(sandbox): clarify section references in guides by @RC-CHN in #6591
• fix: prevent wecom ai bot long connection replies from disappearing by @shijianhuai in #6606
• fix(wecom-aibot): significantly improve streaming readability and speed via add throttling by @Soulter in https://git...

v4.20.1

What's Changed

新增

• 补充 MiniMax Provider。（#6318）
• 新增 WebUI ChatUI 页面的会话批量删除功能。（#6160）
• 新增 WebUI ChatUI 配置发送快捷键。（#6272）

优化

• 优化 UMO 处理兼容性。（#5996）
• 重构 _extract_session_id，改进聊天类型分支处理。（#5775）
• 优化聊天组件行为，使用 shiki 进行代码块渲染。（#6286）
• 优化 WebUI 主题配色与视觉体验。（#6263）
• 优化 OneBot @ 组件后处理，避免消息文本解析空格问题。（#6238）

修复

• 修复创建新 Provider 后未同步 providers_config 的问题。（#6388）
• 修复 API 返回 null choices 时的 TypeError。（#6313）
• 修复 QQ Webhook 重试回调重复触发的问题。（#6320）
• 修复流式模式下 delta 为 None 导致工具调用时报错的问题。（#6365）
• 修复模型服务链接说明文字错误。（#6296）
• 修复 AI 在 tool-calling 模式设为 skills-like 时发送媒体失败的问题。（#6317）
• 修复 Telegram 适配器中 GIF 被错误转成静态图的问题。（#6329）
• 将 Provider 图标来源替换为 jsDelivr CDN 地址，修复部分环境下图标加载问题。（#6340）
• 修复 QQ 官方表情消息未解析为可读文本的问题。（#6355）
• 修复 WebChat 队列异常时流式结果页面崩溃的问题。（#6123）
• 修复子代理 handoff 工具在插件过滤时丢失的问题。（#6155）
• 修复 Cron 提示文案缺少空格及 utcnow() 的弃用警告问题。（#6192）
• 修复 WebUI 启动时 Sidebar hash 导航抖动/定位问题。（#6159）
• 修复启动重试过程中移除已移除 API Key 的 ValueError 报错。（#6193）
• 修复 README 启动命令引用更新为 astrbot run。（#6189）
• 修复 Plain.toDict() 在 @ 提及场景下空白字符丢失的问题。（#6244）
• 修复 provider 依赖重复定义问题。（#6247）
• 修复 Telegram 中普通回复被误判为线程的处理问题。（#6174）

其他

• 调整 astrbot.service 及 CI 配置，升级 GitHub Actions 版本。

---

What's Changed (EN)

New Features

• Added OpenRouter chat completion provider adapter with support for custom headers (#6436).
• Added MiniMax provider (#6318).
• Added batch conversation deletion in WebChat (#6160).
• Added send shortcut settings and localization support for WebChat input (#6272).
• Added local temporary directory binding in YAML config (#6191).

Improvements

• Improved UMO processing compatibility (#5996).
• Refactored _extract_session_id for chat type handling (#5775).
• Improved chat component behavior and uses shiki for code-block rendering (#6286).
• Improved WebUI theme color and visual behavior (#6263).
• Improved OneBot @ component spacing handling (#6238).
• Improved PR checklist validation and closure messaging.

Bug Fixes

• Fixed missing providers_config sync after creating new providers (#6388).
• Fixed TypeError when API returns null choices (#6313).
• Fixed repeated QQ webhook retry callbacks (#6320).
• Fixed tool-calling streaming null delta handling to prevent AttributeError (#6365).
• Fixed model service link wording in docs/config (#6296).
• Fixed AI media sending failure when tool-calling mode is set to skills-like (#6317).
• Fixed GIF being sent as static image in Telegram adapter (#6329).
• Replaced npm registry URLs with jsDelivr CDN for provider icons (#6340).
• Fixed QQ official face message parsing to readable text (#6355).
• Fixed WebChat stream-result crash on queue errors (#6123).
• Preserved subagent handoff tools during plugin filtering (#6155).
• Fixed cron prompt spacing and deprecated utcnow() usage (#6192).
• Fixed unstable sidebar hash navigation on startup (#6159).
• Fixed ValueError in retry loop when removing an already removed API key (#6193).
• Updated startup command to astrbot run across READMEs (#6189).
• Preserved whitespace in Plain.toDict() for @ mentions (#6244).
• Removed duplicate dependencies entries (#6247).
• Fixed Telegram normal reply being treated as topic thread (#6174).

Documentation

• Updated rainyun backup/access documentation (#6427).
• Updated package.md and platform docs, including Matrix and Wecom AI bot documentation.
• Fixed Discord invite link in community docs.

Chores

• Updated PR templates/checklist workflow, repository service config, and automated checks.
• Refreshed repository automation and formatting maintenance, and removed obsolete changelog scripts.

v4.20.0

What's Changed

新增

• 新增俄语翻译（#6081）。
• QQ 官方 Bot 新增文件、语音、视频消息支持（含 WebSocket 模式）（#6063）。

优化

• 优化 QQ 官方 Bot 的流式消息投递可靠性与主动媒体发送能力（#6131）。
• 优化边界场景下 booter 选择逻辑与消息发送工具（#6064）。

修复

• 修复 Dashboard README 对话框锚点导航失效（#6083）。
• 优先使用具名 weekday 的 cron 示例，避免歧义（#6091）。
• 修复插件市场安装后状态未及时刷新的问题（#6124）。
• 修复插件依赖安装逻辑：仅安装缺失依赖（#6088）。
• 移除 Telegram 适配器中已废弃的 normalize_whitespace 参数（#6044）。
• 修复 Windows 本地 skill 文件读取问题（#6028）。
• 修复 Discord pre-ack emoji 配置重启后不持久化的问题（#6031）。
• 统一 WebUI 搜索框清空行为（#6017）。
• 优化插件依赖自动安装流程与 Dashboard 安装体验（#5954）。

---

What's Changed (EN)

New Features

• Added Russian translation support (#6081).
• Added file, voice, and video message support for QQ Official Bot (including WebSocket mode) (#6063).

Improvements

• Improved streaming message delivery reliability and proactive media sending for QQ Official API (#6131).
• Optimized booter selection logic in edge cases and message sending tooling (#6064).

Bug Fixes

• Fixed broken README dialog anchor navigation in the Dashboard (#6083).
• Preferred named weekday cron examples to reduce ambiguity (#6091).
• Fixed plugin market install-state refresh after installation (#6124).
• Fixed plugin dependency installation logic to install only missing packages (#6088).
• Removed deprecated normalize_whitespace parameter in the Telegram adapter (#6044).
• Fixed local skill file reading issues on Windows (#6028).
• Fixed Discord pre-ack emoji config not being persisted across restarts (#6031).
• Unified WebUI search input clear behavior (#6017).
• Improved plugin dependency auto-install flow and Dashboard installation experience (#5954).

What's Changed

• fix: 插件依赖自动安装逻辑与 Dashboard 安装体验优化 by @zouyonghe in #5954
• fix(webui): unify search input clear behavior by @Flartiny in #6017
• fix: Persist Discord pre-ack emoji config across restart by adding missing default key by @Jzjerry in #6031
• fix: improve Windows local skill file reading by @zouyonghe in #6028
• feat: added support for file, voice, and video messages for QQ Official Bot (including WebSocket mode). by @Soulter in #6063
• perf: optimize booter selection for edge cases and message sending tool by @Soulter in #6064
• fix(docs): typo in docker.md & napcat.md by @shudorcl in #6048
• fix(telegram): remove deprecated normalize_whitespace param from telegramify_markdown.markdownify calls by @DOHEX in #6044
• fix: install only missing plugin dependencies by @zouyonghe in #6088
• chore: update dependency and workflow versions by @zouyonghe in #6119
• fix(extension): refresh plugin market install state after install by @letr007 in #6124
• docs: 添加 Astrbook 和玖帕喵社区链接 by @advent259141 in #6135
• [Security] Fix CRITICAL vulnerability: V-004 by @orbisai0security in #6093
• fix: prefer named weekday cron examples by @stablegenius49 in #6091
• fix(dashboard): restore README dialog anchor navigation by @stablegenius49 in #6083
• feat: add Russian translation by @seoeaa in #6081
• perf(QQ Official API): improve streaming message delivery reliability and proactive media sending by @2ndelement in #6131

New Contributors

• @Jzjerry made their first contribution in #6031
• @shudorcl made their first contribution in #6048
• @DOHEX made their first contribution in #6044
• @orbisai0security made their first contribution in #6093
• @stablegenius49 made their first contribution in #6091
• @seoeaa made their first contribution in #6081
• @2ndelement made their first contribution in #6131

Full Changelog: v4.19.5...v4.20.0

v4.19.5

What's Changed

新增

• Lark 适配器支持 CardKit 流式输出（飞书）(#5777)。
• WebUI 已安装插件列表新增筛选与排序功能 (#5923)。

优化

• 启动时后台加载 MCP Server，不阻塞加载流程 (#5993)。

修复

• 部分情况下 MCP 页报错 500 导致查看不了 MCP 服务器 (#5993)。
• 修复 TTS Provider 测试：增加文件大小校验，并补充 MiniMax 空音频检测 (#5999)。
• 修复前端切换到 Chat 后又回到 Welcome 时，页面切换配置未正确持久化的问题 (#5792)。
• 修复 Azure TTS 不支持 84 位订阅密钥的问题 (#5813)。

文档

• 文档仓库迁移：将 AstrBotDevs/AstrBot-docs 内容迁移至 AstrBotDevs/AstrBot (#5960)。

---

What's Changed (EN)

New Features

• Added CardKit streaming output support for the Lark/Feishu adapter (#5777).
• Added filtering and sorting for installed plugins in the WebUI (#5923).

Impprovement

• MCP Server now loads in the background during startup without blocking the loading process (#5993).

Bug Fixes

• Added file size validation in TTS provider tests and MiniMax empty-audio detection (#5999).
• Fixed frontend state persistence when switching from Chat back to Welcome (#5792).
• Fixed Azure TTS support for 84-character subscription keys (#5813).
• Reverted the MCP stdio missing-command error wording change after the previous fix (#5992).

Documentation

• Migrated documentation content from AstrBotDevs/AstrBot-docs into AstrBotDevs/AstrBot (#5960).

v4.19.4

What's Changed

新增

• 企业微信智能机器人支持长连接模式。#5930

New

• Wecom AI Bot supports long-connection mode(Websockets). #5930