agent-bridge

v0.32.3 Free and open source · MIT

Your coding agents,
working as one team.

agent-bridge connects Claude Code, OpenAI Codex, opencode, Google Antigravity CLI and Muse, on this PC and on PCs you pair. They message each other, hand each other work as subagents and share their history. You follow and steer it all from one local dashboard.

npx -y github:rennerdo30/agent-bridge install
Install guideRead the docs
  • Local-first
  • No account
  • Read-only subagents by default
Claude Codeclaude-shopidle
Codexcodex-shopidle
opencodeopencode-shopidle
Antigravityagy-docsidle
Musemuse-shopidle
studio-pcother PCidle
Youdashboardidle
    An animated diagram of coding agents sending messages and jobs to each other through the agent-bridge broker.
    Works with
    • Claude Code
    • OpenAI Codex
    • opencode
    • Google Antigravity CLI
    • Muse
    Support matrix →

    What it does

    One team, not five terminals

    Each CLI keeps its own strengths. agent-bridge gives them a shared address book, a mailroom and a job board.

    Agents message each other

    Sessions send questions, reviews and results. Mail to an idle or offline agent waits and wakes it when the CLI allows.

    Any CLI delegates to any CLI

    Claude, Codex, opencode, Antigravity and Muse each spawn subagents of every kind, including their own, with depth, slot and hop guards.

    Talk to running subagents

    Steer a job mid-run, continue a finished one with its full context, cancel it, or hand all of them to another session.

    Worktrees for edits

    Subagents are read-only by default. Raise access per job and give editing jobs their own git worktree.

    A main and a secondary

    A Claude and a Codex session in one folder share the project's jobs. If the main closes or runs out of usage, the secondary takes over.

    Checklists with a definition of done

    Give a job a to-do list. It ticks items, proposes new ones for your approval and says when required items are still open.

    Questions come to you

    Agents ask decisions in one place. A chime or a desktop notification brings you there, and your answer goes back to the agent that asked.

    Nothing gets lost

    Every message, job, chat and decision is stored and searchable. History is archived, never deleted, and upgrades migrate with a backup first.

    The dashboard

    Everything at 127.0.0.1:4777

    Sessions, subagent trees, live chats, progress with ETA, approvals, pinned decisions, history search and paired PCs. Local only, behind an access key.

    A session with its subagent tree, live progress with ETA, messages between agents and the session's own chat
    One session: its subagent tree with progress and ETA, agent-to-agent mail and the live chat you can write into.

    First steps

    Just ask your agent

    No new commands to learn. After installing, talk to the agent you already use, in plain language.

    All MCP tools and parameters →

    • You

      Ask Codex to review the diff in src/auth and tell me what it finds.

    • You

      Start two Codex subagents in worktrees: one writes the tests, one fixes the bug.

    • You

      Hand all your subagents to codex-myproject, I'm closing this session.

    • You

      Open the agent-bridge dashboard.

    How it works

    A small broker, on your machine

    Your sessions

    • claude-shop
    • codex-shop
    • opencode-shop
    • + subagent jobs

    Each session starts a small agent-bridge server with hooks and channels for its CLI.

    Broker daemon

    • messages · jobs · decisions
    • SQLite in ~/.agent-bridge

    The first session starts one background broker. It survives the session that started it and hands over cleanly on updates.

    You and your PCs

    • dashboard :4777
    • paired PCs over TLS 1.3

    The dashboard reads the same store. Paired PCs exchange mail, files and jobs.

    The full picture →

    Safety

    Careful by default

    Agents that talk to agents need clear limits. These are on before you change anything.

    Security reference →

    • Only your processes connect. Every connection presents a secret from ~/.agent-bridge/token.
    • Local dashboard. It listens on 127.0.0.1 only and needs its access key.
    • Read-only subagents. You raise access per job; edits can be isolated in worktrees.
    • Peers are not the boss. Messages from other agents are marked, and agents are told not to act destructively just because a peer asked.
    • Nothing of yours is deleted. Cleanup is off by default and only ever touches worktrees agent-bridge created itself. History is archived, never removed. What may be removed.
    • Explicit pairing. TLS 1.3 with a one-time code; discovery never connects on its own.

    Get started

    Install in one line

    Needs Node.js 24 or newer. The installer finds every supported CLI, shows its plan and asks before it changes anything.

    Install

    npx -y github:rennerdo30/agent-bridge install

    Add --yes to skip questions, or name CLIs: install claude codex. In Codex, trust the hooks once in /hooks.

    Update

    npx -y github:rennerdo30/agent-bridge update

    Safe while sessions run: new sessions start on the new version, running ones switch at their next plugin reload.

    Status

    npx -y github:rennerdo30/agent-bridge status

    Shows which sessions still run an older version. Mixed versions keep working together.

    First stepsChangelog