| How you set it up | Download the app for Mac, Windows or Ubuntu. Nothing to host. Murage opens the one install line for each engine you pick | Run a Node.js server (Node.js 24.11+ and pnpm 9.15+) from an install script or a git clone, with an embedded PostgreSQL database locally |
|---|
| How you give it work | Talk to your Chief of Staff in plain English, in the app or from your chat apps. It proposes the team and hands out the work | Issues and tickets tied to projects and goals. Agents pick them up when assigned or on scheduled heartbeats |
|---|
| An org chart of your agents | Yes, built for you. The Chief sets up teams with team leads for each job, shown on the Team map, and you can reshape it | Yes, designed by you: roles, titles, reporting lines and permissions |
|---|
| Memory across jobs | Yes. Each bot keeps a notebook, plus managed memory stored on your computer | On its roadmap (“Memory / Knowledge”), not shipped yet. Agents resume a task’s context across heartbeats |
|---|
| Voice and chat apps | Call any bot in the app. Your Chief answers in Telegram, Slack and Discord, and a voice message gets a voice reply | Not mentioned in its docs |
|---|
| Ready-made teams and skills | 68 ready-made teams and 2,237 skills, each scanned by Skill Guard for risky instructions | Import and export whole companies (agents, skills, projects, routines), plus a Skills Manager and Skills Store. No ready-made count published |
|---|
| Scheduled work | Routines: scheduled jobs for any bot or team, while your computer is on | Heartbeats: agents wake on a schedule, check their work and act |
|---|
| Hard spend limits | No. Usage and cost are tracked per bot, but there are no hard caps yet | Yes. Budgets per agent, project and goal with hard stops: overspending pauses the agent and cancels its queued work |
|---|
| Runs OpenClaw | Not yet. OpenClaw is coming as a Murage engine | Yes, as an agent on the team |
|---|
| Asks before acting outside, by default | New bots start on Ask. On Full access, a bot stops before messaging anyone new, posting publicly, paying, or deleting outside its folder, unless you have allowed it for that task. Replies in an existing conversation go ahead on Auto and Full access. No limits turns the stop line off. On Claude Code and Codex, the stop line also covers tools from MCP servers you add. On Hermes and other ACP engines, Ask and the stop line cover only what the engine asks Murage about: Hermes asks before shell commands and file edits, but not before MCP or connected-app tools. | Approval workflows, execution policies with review stages, and audit logs that you set up. Per-action defaults aren’t stated |
|---|
| Which AI | The engine you choose, per bot: Claude Code, Codex, Hermes, Grok Build, Cursor, Google Antigravity, Kimi, Qwen, Droid, OpenCode, Pi, Fuigo (built in), local models (Ollama, LM Studio) and any OpenAI-compatible API (chat only). OpenClaw is coming. | Claude Code, Codex, Cursor, Bash, HTTP webhooks and OpenClaw, plus adapter plugins |
|---|
| On your phone | The web app, plus your Chief in Telegram, Slack and Discord. iPhone and Android apps coming | A responsive web UI in your phone’s browser. No native app |
|---|
| Keeps working with your laptop closed | Not today. It runs while your computer is on; Murage Cloud isn’t open yet | Yes, if you run it on a server |
|---|
| Price | Free app. Your AI provider bills as usual | Free. The agents’ own plans bill as usual |
|---|
| Open source | Yes, AGPL-3.0 | Yes, MIT |
|---|