LangBot v4.11.8

LangBot 是一个开源的 AI 聊天机器人与智能体平台，支持多种即时通讯平台、大语言模型、知识库和插件，帮助你构建、部署和管理 AI 助手。
LangBot is an open-source AI chatbot and agent platform that connects messaging platforms, large language models, knowledge bases, and plugins to help you build, deploy, and manage AI assistants.

LangBot v4.11.8

Share prompts, Skills, MCP servers and plugins to LangBot Space from the WebUI, get alerted when bots or models keep failing, and pick the sandbox backend per Workspace. The Docker image is 60% smaller.

本版本支持从 WebUI 把提示词、Skill、MCP 服务器和插件分享到 LangBot Space，新增工作区故障告警，可以按工作区选择沙箱后端；Docker 镜像缩小约 60%。

Upgrade note / 升级提示：本版本包含新的数据库迁移（0047、0048），升级前请备份 data/ 目录（使用 PostgreSQL 时同时备份数据库）；降级需要用这份备份恢复。This release adds database migrations 0047 and 0048. Back up data/ (and PostgreSQL, if used) before upgrading; rolling back needs that backup.

Highlights

• LangBot Space sharing / 分享到 LangBot Space：在 WebUI 里直接把对话提示词、Agent 提示词、Skill、远程 MCP 服务器和本地安装的插件提交到 LangBot Space。提交前会扫描凭据：提示词里有密钥会被拦下，远程 MCP 的鉴权头和参数会替换成占位符，不会离开本机。Pipeline 和 Agent 的提示词输入框上方新增提示词库，可以搜索、预览、替换或追加，并与 space.langbot.app 的收藏同步。Share prompts, Skills, remote MCP servers and locally installed plugins to LangBot Space from the WebUI, with credential scanning; browse and favorite community prompts from the prompt editor.
• Workspace alerts / 工作区告警：机器人掉线、流水线或模型连续失败时，通过邮件、LangBot Space 或 Webhook 通知管理员，阈值可调，恢复后自动关闭。邮件配置移到顶层 email 段（SMTP 或 Resend），旧的 workspace.invitations.email 会在启动时自动迁移。Alerts for offline bots and repeated pipeline/model failures via email, LangBot Space or webhooks; email settings move to a top-level email section and are migrated automatically.
• Workspace sandbox / 工作区沙箱：每个工作区可以选择 Docker、E2B 或 nsjail 作为 Box 后端，并在管理员上限内设置资源限制；修改只影响新任务。Box 支持网络文件传输，文件可以在 Core、Runner 和沙箱之间传递，并带有校验缓存。Pick Docker, E2B or nsjail per Workspace within administrator limits; Box can now move files between Core, Runners and sandboxes over the network.
• Skills / 技能：Skill 以版本化的包保存（含附带文件），可以导出、分享到 Space。Skills are stored as versioned packages with their assets and can be exported and shared.
• Smaller Docker image / 更小的 Docker 镜像：rockchin/langbot 从 768 MB 降到 308 MB（gzip，amd64）。SeekDB、Milvus、Valkey Search 改为可选：Docker 用户使用 <tag>-seekdb / -milvus / -valkey 镜像，pip/uvx 用户安装 langbot[seekdb] / [milvus] / [valkey]。使用默认的 Chroma、Qdrant、pgvector 不受影响。The default image drops from 768 MB to 308 MB. If you use vdb.use: seekdb, milvus or valkey_search, switch to the matching image tag or extra.
• Agent runner lifecycle / Agent Runner 生命周期：取消、超时、失败和断连时保留已产生的 token 用量；新增进度和取消结果、可确认的 steering 投递、重试与取消确认（SDK 0.7.25）。Runner runs keep partial usage through cancellation and disconnects, with progress/cancel results, acknowledged steering and retry (SDK 0.7.25).
• LocalAgent：模型返回空回答时切换到下一个备用模型，没有备用模型时报错，不再静默无回复；发给模型的消息只保留一条开头的 system 消息并保证角色交替，修复 vLLM/SGLang 上的 Qwen3.5/3.6 挂知识库后报 Invalid request parameters。Empty model answers now fall back or fail visibly; Qwen3.5/3.6 on vLLM/SGLang no longer rejects LocalAgent requests with a knowledge base attached.
• Fixes / 修复：n8n 流式工作流报错时 Runner 不再报告成功；插件 call_tool 结果序列化为 JSON；出站 URL 校验拒绝所有非公网地址（包括 100.64.0.0/10，含云厂商元数据地址）；DifyAgent 不再在请求日志中记录表单 token 和签名 URL；LangBot Cloud 在服务端临时故障时不再把用户登出，可上传的插件详情页增加上传按钮。
• SDK：插件 SDK 已并入 LangBot 主仓库 sdk/ 目录并独立发版；langbot-plugin 由 0.7.24 固定到 0.7.25。The plugin SDK now lives in this repository under sdk/; Core pins langbot-plugin==0.7.25.

Release verification

• Release commit 9301e0ea3eec6bc3a57ef85475557c6b807ae80e. It differs from the fully tested candidate 7e2668682 only in the version number; all pull-request checks passed on that candidate, including unit tests (Python 3.11/3.12/3.13), PostgreSQL contract tests, Core with the in-tree SDK, E2E startup, Core smoke and regression tools. On the release commit itself, the uvx install passed a smoke run (plugins, model, LocalAgent chat with a tool call, commands, !version = v4.11.8) and the Docker image started cleanly.

• Before tagging, the candidate was installed twice on fresh data directories and run through routine operations:

  • Docker (docker compose with the plugin runtime and Box, image built from the release commit; SQLAlchemy 2.0.46);
  • uvx (uvx --from <wheel> langbot, dependencies from PyPI: SDK 0.7.25, SQLAlchemy 2.1.8).

  Both passed the same checks: account setup and WebUI; Marketplace install of LocalAgent, EssentialCommands, WebSearch, LangRAG and HelloPlugin with permission confirmation, enable/disable per pipeline and uninstall; model provider and model test; a LocalAgent pipeline over HTTP Bot, including a web-search tool call; !help !cmd !func !plugin !version; knowledge-base ingestion and retrieval; creating a Skill and having the agent use it; 8 simultaneous conversations on one model (no token-count rejections); and data, plugins, knowledge base and Skill intact after a restart. Docker also ran a command in a Box sandbox container.

• Windows: the release wheel was installed with uvx on Windows 11 (code page 936, Python 3.11, SDK 0.7.25, SQLAlchemy 2.1.8) and passed the same checklist with a local mock model, plus a plugin printing Chinese and emoji (runtime stayed connected) and a first LocalAgent conversation with tokenizer downloads blocked (2 s, no count_tokens timeout).

• Database migrations: databases created by v4.11.7 (revision 0046), v4.10.11 (0023) and v4.10.6 (0008, before workspaces) were upgraded to this release on both SQLite and PostgreSQL. Each upgrade reached revision 0048 and started twice cleanly; no rows were lost, SQLite integrity and foreign-key checks passed, and the seeded provider, model, bot and pipeline read back. Copies of three real SQLite databases (revisions 0042, 0040 and 0008, up to 33 MB) also upgraded cleanly. Upgraded schemas match a fresh install, apart from index naming differences on databases that started before 0024, which v4.11.7 has as well.

Contributors

@RockChinQ @huanghuoguoguo @Tynwink2000 @goranai

New Contributors

• @goranai made their first contribution in #3029

Full Changelog: v4.11.7...v4.11.8