raysonmeng

raysonmeng / agent-bridge

Public

A local bridge for bidirectional collaboration between Claude Code and Codex. 连接 Claude Code 与 Codex 的本地实时协作桥接工具。

agent-collaboration claude-code cli codex developer-tools
36
3
100% credibility
Found Mar 25, 2026 at 36 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

AgentBridge is a local tool that connects Anthropic's Claude Code and OpenAI's Codex CLI for real-time bidirectional messaging, enabling collaborative AI-assisted coding workflows.

How It Works

1
🔍 Find AgentBridge

You hear about a handy tool that lets two AI coding buddies chat and team up on your coding projects right on your computer.

2
📥 Set it up

Download the tool and prepare it on your machine so it's all ready to connect your AIs.

3
🚀 Launch with first AI

Open your main AI coding helper and link it to the bridge – it starts a background helper to wait for the second AI.

4
🔌 Bring in the second AI

Open another window for your second AI coding helper and connect it – now both AIs can see and talk to each other!

5
💬 Give them a job

Tell them what to build or fix, and watch as they message back and forth, reviewing ideas and making changes together.

🎉 Teamwork magic

Your code gets smarter and faster, with both AIs collaborating like pros to deliver great results.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 36 to 36 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is agent-bridge?

Agent-bridge is a TypeScript tool built with Bun that creates a local bridge for bidirectional messaging between Claude Code and Codex in the same coding session. It solves the silos between AI agents by letting Claude review and reply to Codex outputs via MCP channels or tools, while Codex gets injected user turns from Claude—all running locally without hosted services. Developers get real-time collaboration: launch via Claude's dev channels, attach Codex TUI to a proxy port, and watch agents debate code changes.

Why is it gaining traction?

It stands out as a bridge AI agent for local setups, enabling human-in-the-loop multi-agent workflows without cloud dependencies or generic orchestrators. The hook? Agents built it themselves—Claude and Codex collaborated via the bridge to ship the code—proving seamless turn-taking, loop prevention, and message filtering with markers like [IMPORTANT]. Config via env vars like CODEX_WS_PORT keeps it lightweight for local GitHub Copilot alternatives.

Who should use this?

AI tooling tinkerers experimenting with agent bridges, backend devs chaining LLMs for code review workflows, or researchers testing multi-provider agent collaboration. Ideal for those hitting Codex git sandbox limits—let Claude handle commits while Codex implements—without multi-session complexity.

Verdict

Try it for local agent-bridge proofs-of-concept if you're into bridge AI agent hacks; solid docs, tests, and bilingual README make setup straightforward despite 36 stars and 1.0% credibility score. Still experimental with single-thread limits, so not production-ready—use for weekend spikes on local GitHub actions runners.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.