botmux 文档
  • English

      • Multi-bot Collaboration

      Multiple Lark bots can run on the same machine, and each bot can be bound to a different CLI. Add them to the same group and you can have different models "duke it out" against each other.

      Routing rules

      • Just you and one bot in the group: no @ needed, it responds automatically (under the default @ policy this requires both "1 human + 1 bot"; a multi-human group needs an @ even with a single bot).
      • Multiple bots / multiple humans in the group: use @mention to specify the recipient.
      • @botA @botB /t <prompt>: have each @-mentioned bot open its own independent session on this message, each running on its own (both bots anchor on your /t message and share the same thread, but the sessions are independent).

      Letting bots know each other

      botmux discovers bots in the current Lark group via the group bot roster by default. Cross-deployment recognition has two paths, both entering the same team-trust gate and skipping /grant: one is the federation in the Dashboard "Teams" panel (invite someone else's deployment into the same team, create cross-deployment groups); the other is central-platform team-sync (the platform pushes down teams and groups). A plain /group or the Dashboard "New Group" flow creates an ordinary same-deployment group that uses neither path and carries no cross-deployment team trust.

      For handoffs, have the model check:

      • botmux bots list: lists bots in the current group, capability tags, openId, and mentionable.
      • The <available_bots> block in the message context: hints which bots can be relayed to.

      After that, a bot can explicitly @ the other from its own session with botmux send --mention <the other's openId>, triggering the relay.

      A hard fact about cross-bot collaboration: if you don't --mention the other bot, it won't be triggered at all.

      Team scope: cross-machine discovery + create-group / invite

      Once connected to the central platform, you can discover agents on other people's machines within the same team by specialty (even ones not yet in your group), and pull them in to collaborate. Three commands, each with a distinct job:

      # Discover: list same-team, opted-in agents (use --team when this machine is in several teams)
botmux bots list --scope team [--team <teamId>]

# New group: pull discovered agents (by appId) + their owners into a platform-created focused group
botmux create-group --team <teamId> --agent <appId> [--agent ...] [--name "Group name"]

# Add to an existing group: add agents + their owners into a group you're already in (owners always included)
botmux bots invite --chat <chatId> --team <teamId> --agent <appId> [--agent ...]

      Each agent from bots list --scope team carries: appId (use it to invite), name, specialties (an array of tags), mentionable, online, owner, machineId/machineName.

      Key points:

      • Opt-in gate: the discovery list only contains agents that joined the team — i.e. their owner explicitly added them via the platform's "Manage bots". If an agent doesn't show up, its owner most likely hasn't added it yet.
      • specialties / mentionable / online are self-reported by the agent — for selection only, not a trusted credential.
      • appId only, no @ needed: precisely because you can't @ someone else's bot before it's in your group, team discovery/create-group/invite all go by appId + platform auth, sidestepping "can't click what you can't see". Authorization (team membership, opt-in) lives entirely on the platform; the CLI makes no such judgment.
      • bots invite target group requires: you are already in it and the platform bot is in it. When the platform bot is absent, botmux automatically uses a local bot already in the group to add it, then retries — usually invisible; only when the group requires owner approval to add bots, or you have no local bot in that group, will it ask you to add it manually.

      Don't confuse the three add paths: create-group --team (cross-team new group) / bots invite (add same-team people into a group you're already in) / Lark /invite (add a same-tenant bot within the current group).

      Typical scenarios

      • Code review: have Claude Code and Codex review the same MR together, picking holes in each other's views when they disagree.
      • Design review: two AIs each propose a solution and find flaws in each other's.
      • Write/review separation: one writes, one reviews, iterating to convergence.

      Configuration is minimal: one Lark app per bot, add them to the same group, and the processes are fully isolated.

      Related: use Roles & Teams to give each bot a persona and capability tags; use One-click session group to add multiple bots to a new group in one sentence; use Session Relay to move a session into a collaboration group.