Important
This repository has moved. The plugin SDK, lbp CLI, Plugin Runtime and Box Runtime now live in the sdk/ directory of the LangBot repository. Please open issues and pull requests there.
This repository is kept for history only (up to v0.7.24) and no longer receives changes. The PyPI package is unchanged: pip install langbot-plugin. New versions are released from LangBot with sdk/vX.Y.Z tags.
This repository is the shared infrastructure powering LangBot's plugin and
sandbox subsystems. It is published to PyPI as the single
langbot-plugin package and ships
three things:
- Plugin SDK & CLI (
lbp) — Python APIs, base classes and thelbpcommand for scaffolding, building, debugging and publishing plugins. - Plugin Runtime (
lbp rt) — the host process that discovers, installs and runs plugins, bridging them to LangBot over stdio or WebSocket. - Box Runtime (
lbp box) — the code-sandbox service backing LangBot's Box subsystem, executing untrusted code via Docker / nsjail / E2B backends. An explicit, non-sandboxedhostbackend is available for trusted local development only.
LangBot depends on this package as the pinned langbot-plugin==<x.y.z> in its
pyproject.toml; the canonical version lives in this repo's pyproject.toml.
The Runtime component (
src/langbot_plugin/runtime/) is licensed separately under AGPL; everything else in this repository is Apache 2.0. Seesrc/langbot_plugin/runtime/README.md.
pip install langbot-plugin
# or, recommended
uv tool install langbot-plugin
lbp --help| Command | Purpose |
|---|---|
lbp init <name> |
Scaffold a new plugin project |
lbp comp <Type> |
Generate a component (Command / Tool / EventListener / KnowledgeEngine / Parser / Page) |
lbp run |
Run / remote-debug a plugin against a running LangBot |
lbp build |
Package the plugin into a distributable zip |
lbp publish |
Publish to the LangBot Plugin Market |
lbp login / lbp logout |
Authenticate with your LangBot account |
lbp rt |
Launch a standalone Plugin Runtime (control 5400, debug 5401) |
lbp box |
Launch a standalone Box sandbox runtime (default port 5410) |
lbp ver |
Print the CLI / package version |
Plugins extend LangBot through six component types, scaffolded with
lbp comp <Type>:
- Command — user-triggered actions (e.g.
!weather tokyo) - Tool — LLM-callable functions for AI agents
- EventListener — handlers for message-pipeline events
- KnowledgeEngine — custom knowledge-base retrieval for RAG
- Parser — custom message / content parsing
- Page — custom web page embedded in the LangBot admin panel
The SDK provides Runner component definitions, protocol entities, runtime
dispatch, run-scoped Host API helpers, and low-level daemon relay primitives
used by runner/plugin packages that need an outbound user-side runtime process.
This includes run-scoped pull APIs on RunnerAPIProxy (history, events, state,
run ledger) and steering_pull for absorbing follow-up user input into an
active run when the runner declares capabilities.steering in its manifest.
The SDK does not provide a managed, generic remote Runner product: external
harness execution policy, process launch details, authentication, workspace
selection, and provider-specific behavior belong to the runner/plugin package or
a deployment-owned remote daemon or managed agent platform.
LangBot Core 通过统一的 ExecutionAPI facade 暴露 72 项核心业务能力。插件与 Agent Runner 统一经由该 facade 与 SDK 入口访问。
- 通用执行入口:
api.execution.invoke(operation_id, **arguments)与api.execution.stream(...)。 - Box 沙箱代理扩展:
LangBotAPIProxy现已提供@run_scoped的bind_box,import_box_attachments,export_box_files。 - 完整清单与作用域契约: 详见 docs/execution_api.md。
- Plugin development tutorial — https://langbot.app/docs/zh/plugin/dev/tutor
- Debugging the Runtime / CLI / SDK — https://langbot.app/docs/zh/develop/plugin-runtime
- Dev environment setup — https://langbot.app/docs/zh/develop/dev-config
- ExecutionAPI 操作一览与作用域说明 — docs/execution_api.md
Apache 2.0, except the Plugin Runtime (src/langbot_plugin/runtime/) which is
AGPL. See LICENSE and the
Runtime README.