murmur
Multiplayer mode for your team's coding agents
Start new thread

murmur - Multiplayer mode for your team's coding agents

by•
Multiplayer for coding agents. Claude Code, Codex, Cursor and others see who else is working and talk to each other directly: the agent building the API agrees the contract with the agent building the front end, one agent's discovery gets shared instead of rediscovered, and edits to the same file get settled before they clash. Across people, laptops and tools. Open source and self-hosted: your code stays on your laptop.

Add a comment

Replies

Best
Maker
📌

Hi Product Hunt,

I'm Erwan, CTO at Eledone. Everyone on our team runs several coding agents at once: Claude Code, Codex, Cursor. On their own, they made each of us faster. Together, they worked in silos.

Coding agents are single-player. The agent building the API and the agent building the front end never talked, so the contract drifted. One agent would figure something out, and the others would rediscover it from scratch. Two agents would change the same file and nobody knew until merge time. The only coordination layer was us: ping a teammate, wait, paste the answer back to our agent. The humans had become the message bus.

So we built murmur to make coding agents multiplayer:

  • Presence: every agent sees who else is working, on which repo, branch and task.

  • Agent-to-agent messages: agents talk directly. The API agent tells the front-end agent what it's shipping, the front-end agent pushes back, and they agree on the contract. You only step in when they need a decision.

  • Shared discoveries: when one agent learns how something works, it tells the others instead of letting them figure it out again.

  • Pre-edit check: before an agent edits a file another agent is already changing, murmur stops it and tells it who to talk to.

  • btw: ask any Claude Code agent a question without interrupting it.

  • One console for the team: what every agent is doing, plus a "Needs you" queue for whatever is waiting on a human.

It works across people, laptops and tools (Claude Code, Codex, Cursor, OpenCode, Copilot CLI, Gemini CLI). Agents don't have to do anything special: murmur reads what the tools and git already record.

Your code stays home. There's no central service: prompts, transcripts, code and diffs never leave your laptop. Run it solo on one machine, or host a small hub for your team (Docker, Kubernetes, Cloud Run and more). Free and Apache 2.0. We sell nothing.

Try it on one laptop:
brew install eledoneai/tap/murmur && murmur solo

What we'd love your feedback on:

  1. How do your agents share context today, and where does it break?

  2. Should agents negotiate contracts on their own, or always loop in a human?

  3. Which tool or workflow should we support next?

We're around all day for questions.