Open-source AI agent from the Agentic AI Foundation — a native desktop app, CLI and API that runs tasks locally with 70+ MCP extensions.
Summary
A browser canvas where Claude Code, Codex, Kimi and other coding agents work the same repository at once without overwriting each other.
Screenshots
Description
Murmell is a shared canvas for running several coding agents against one repository at the same time. Each agent gets its own live terminal pane on a single browser canvas, so instead of juggling tabs or worktrees you watch the whole crew work side by side — and so do your teammates, since the canvas is collaborative rather than per-seat.
The file-claiming model. The reason parallel agents normally go wrong is that two of them edit the same file and one silently loses. Murmell enforces one holder per path: before an agent writes, it must claim the file, and while the claim is held no other agent can write there. Conflicting writes are snapshotted before they land, so a collision costs you a review rather than a lost change. This is the product's actual technical claim and the thing that distinguishes it from simply opening four terminals.
Agents supported. Claude Code, Codex, Kimi, OpenCode, Hermes, Cursor and Gemini, run on Murmell's machines rather than yours. You can bring an existing subscription instead of metered API keys.
Pricing. Solo is $39/month for one project and five agent windows. Pro is $69/month for three projects, ten agent windows and a shareable canvas. Builder is $149/month for unlimited projects and windows with team ownership. All tiers include a seven-day free trial, and Pro and Builder bundle $60 of Claude credit.
Who it is for. Teams that have already moved past one-agent-at-a-time and are hitting coordination problems — merge churn, agents undoing each other's edits, no shared view of what is running. It is hosted only; there is no self-host option, so the repository and the execution both live on Murmell's infrastructure, which is worth weighing for private code.
Reviews
Similar App Suggestions
A shared memory layer for AI coding tools — tell Claude Code something once and Cursor, Codex and Claude Desktop already know it.
Free, open-source macOS menu bar app that shows live status for every parallel Codex task, so you can see which agent needs you without reopening Codex.
Watch and approve Claude Code, Codex and OpenCode sessions running on your Mac from an iPhone, with live transcripts and push notifications.
Local-first memory hub that reads your existing AI chat history and gives every agent you use the same long-term context.