botmux 文档
  • English

      • botmux

      Turn a Lark topic group into a remote control for AI coding CLIs. One message launches a dedicated coding session.

      botmux is a bridge: a persistent daemon listens to Lark messages and automatically launches a dedicated AI coding CLI process (Claude Code / Codex / Cursor / Gemini / OpenCode / Antigravity, etc.) for each new topic. It renders terminal output into Lark streaming cards in real time, and provides an interactive Web Terminal. Phone, computer, and Lark stay in sync — wherever you are, your coding session follows.

      Project: https://github.com/deepcoldy/botmux ｜ npm: npm install -g botmux

      Design philosophy: not an SDK wrapper — bridge the CLI directly

      botmux does not reimplement agent capabilities; it bridges directly to existing AI coding CLIs. Memory, context management, tool calls, permission systems, plan mode, / commands, the MCP ecosystem — these are all capabilities the CLIs themselves are iterating on rapidly. botmux chooses to stand on top of that evolution rather than rebuild a parallel copy. Every time a CLI ships an upgrade, botmux benefits automatically with zero adaptation.

      At the same time, botmux uses structured prompt injection (XML tags) inside the daemon to separate user content from system instructions before feeding it to the CLI — this is the prompt format that models reliably handle best. But everyday users neither need to nor should hand-write XML; just send plain language as usual, and botmux handles the wrapping for you.

      Core advantages

      Compared to approaches like OpenClaw that are "rebuilt on top of an Agent SDK":

      FeaturebotmuxSDK-based approach
      Underlying architectureBridges directly to the full CLI processRebuilt on top of an Agent SDK
      CLI capabilitiesFull runtime (hooks / memory / plan mode / Skill / / commands / MCP)A subset of the SDK API; missing features must be added manually
      CLI upgradesBenefit automatically with zero adaptationMust track SDK version changes
      Memory / contextReuses the CLI's built-in memory directly, enhanced as the CLI iteratesMust build your own, duplicating the CLI's native capabilities
      Multiple CLIsOne-click switching among 6+Bound to a single SDK
      Web TerminalFully interactive terminal, synced across three surfacesUsually read-only output
      Multi-bot collaborationMultiple bots in one group with @mention routing, process isolationUsually a single bot
      Direct terminal accesstmux attach into the process, identical to localCannot operate the underlying terminal

      Highlights

      • Streaming cards — One live-updating Lark card per conversation turn, with terminal output rendered as Markdown
      • Interactive Web Terminal — Not just viewing: operate the CLI directly in the browser; mobile has a floating shortcut toolbar
      • Multi-bot collaboration — Put multiple bots with different CLIs in one group, @ whoever you want to work, and have Claude Code and Codex review code together
      • Persistent tmux sessions — Restarting the daemon doesn't interrupt CLI processes
      • Session adopt — Adopt a CLI running in a local tmux into Lark with one click, and continue on another device
      • Scheduled tasks — Configure recurring tasks in natural language; they resume in the original topic when due
      • On-Call Mode — Anchor a group to a project so anyone in the on-call group can @ for instant answers

      ➡️ Next: 5-minute quick setup