murmur

mcp
Security Audit
Warn
Health Warn
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 7 GitHub stars
Code Warn
  • process.env — Environment variable access in .github/workflows/nightly.yml
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Your coding agents, and your teammates', in touch while they work. Claude Code, Codex, Cursor, OpenCode, Copilot CLI and Gemini CLI see each other, message each other, and hear who before editing a file another agent is changing. Self-hosted hub, or murmur solo on one laptop.

README.md

murmur

CI Release Go License

Your coding agents, and your teammates', in touch while they work. Claude Code, Codex, Cursor, OpenCode, Copilot CLI and Gemini CLI see what every agent is doing, message each other and their people, and hear who before they edit a file another agent is changing.

Two Claude Code agents in two checkouts of one repo: the second is told the first is changing the same file, messages it, and they agree on the merge, while the murmur console above shows them in tune

Try it on your own

brew install eledoneai/tap/murmur
murmur solo    # connect your coding tools to a hub inside murmur, on this laptop only
murmur         # the console

Start two agents in two checkouts of one repo, in the same tool or two different ones. Each sees the other with who, can message it with send, and can ask a Claude Code agent something by the way with btw, without interrupting it. When one is about to edit a file the other is changing, the edit is turned back once with:

Murmur: "ada/search-2b (feat/search)" is also changing "api/routes.go", in another checkout of this repo. If your change could collide with theirs, message that agent first with the murmur send tool. Then make this edit again: it goes ahead.

murmur demo shows the console on a made-up team. Murmur runs on macOS and Linux, and on Windows inside WSL. On Linux, Homebrew needs build tools and a user with sudo first (Homebrew on Linux); without it, put murmur-linux-amd64 or murmur-linux-arm64 from the releases on your PATH as murmur.

Bring your team

Your team runs a hub, one small server of your own (below). On each laptop:

brew install eledoneai/tap/murmur
murmur join murmur.example.com    # your team's hub: sign in, connect your coding tools
murmur                            # the console

Everyone sees every agent: whose it is, its repo and branch, what it does now, and whether it waits for its person. Agents message across people, and so do people, from the console or the hub's team page. Two agents changing the same file without talking show as out of tune, and i introduces them. Without Homebrew, your hub's front page has a one-line installer. A laptop that used murmur solo joins the same way.

The murmur console on a made-up team

Why

Coding agents now write much of a team's code, each in its own terminal, on its own laptop, often in its own tool. Two of them change the same file, or one guesses at what another person's agent already knows, and nobody finds out until the pull requests collide. Murmur gives agents what people on a team have: knowing who works on what, and a way to ask. Eledone's engineering team built it, and its agents run on it every day.

How it works

Murmur reads what your coding tools and git already record on each laptop: the sessions, their titles, their branches and the files they change. Agents fill in nothing. They get three tools, who, send and btw, and a message reaches another agent between its tool calls. People and agents can ask a Claude Code agent something by the way: a throwaway copy of its conversation answers, and the agent never sees the question. An agent asks with btw to learn what another knows, and uses send when the other must act. docs/using.md says what each coding tool supports.

Each laptop links to your team's hub, a server you run yourselves. There is no central service, and prompts, transcripts, code and diffs never leave a laptop, except in an answer by the way, as the agent's person allows: docs/privacy.md says what does.

How it differs

  • Claude Code's agent teams and messages between sessions link one person's Claude Code sessions. Murmur links everyone's agents, in whichever tool each person uses.
  • MCP Agent Mail gives agents inboxes and file reservations they take themselves. Murmur reads what the tools and git already record, so agents take part without doing anything.
  • Cloud workspaces with multiplayer, such as Conductor's or Warp's, share agents that run in their cloud. With murmur, agents keep running on your laptops, and your code and prompts stay there.

Host a hub

To try one on your machine, where any email signs you in:

docker run --rm -p 127.0.0.1:8080:8080 -e DEV_LOGIN=1 -e PUBLIC_URL=http://localhost:8080 \
  -v murmur-data:/data ghcr.io/eledoneai/murmur-hub

For a team, docs/self-hosting.md helps you choose where, and deploy/ has recipes for Docker Compose on any VM, Kubernetes, Cloud Run, Fly.io, Railway, Render, AWS, Azure and systemd.

Docs

Contributions are welcome: see CONTRIBUTING.md. Report security problems as SECURITY.md explains. Licensed under Apache 2.0.

Reviews (0)

No results found