One Maurice. Every door.

One Maurice, even on its own

1. ① One Maurice, on one computer. It already runs your tools.

   Installed on your computer, Maurice runs the tools you have compiled. It needs no server and no other Maurice. But it is not locked in: you choose who can reach it, through which door, and to do what.

Every door

1. ② Reachable over HTTPS, with a token limited to what you open.

   For an outside service to reach your Maurice, you give it an address and a token.

   • A scope: one workflow or one application, nothing else.
   • A duration and a quota, set by you.
   • The token is shown only once and can be revoked in one click.

   Your applications also get a REST API, with keys limited to what you expose.

2. ③ Your teams and your customers write to it from their messaging app.

   Telegram, Slack, WhatsApp, Discord, Google Chat, Signal: people write to Maurice where they already are, without installing anything. Maurice routes each request to the right tool, with the permissions of what you opened. For your site or your product, an embeddable chat does the same in a web page.

3. ④ Your AI agent only sees what you open to it.

   Grok Bot, Hark, Claude Code, Codex, Cursor, or any MCP client: the assistant you already use works with Maurice. Muse and Dots connect the same way, wherever they are available.

   • Only three moves: search, get an explanation, run. A call outside what you opened is refused.
   • It prepares, you decide: a sensitive step is approved in your own session, never by the agent.
   • It stays a guest: connecting it gives it neither your keys, nor your vault, nor an admin role.

4. ⑤ Your machines talk to it over MQTT.

   Sensors, controllers and connected devices send their readings to Maurice and receive its answers over MQTT. Each key only reaches its own topics.

One log

1. ⑥ One tool, the same permissions, the same log.

   Whatever the door, the same tool answers, with the same rules. What you expose is defined once. Every run goes into the same log, with its door, its result and its cost.

The swarm, with permissions

The Maurice on your computer, a colleague's, the team server's and the company's can work together. When one needs what another can do, it hands it the request.

1. ⑦ Several Maurices form a swarm. Every link is approved by the receiver.

   A Maurice decides who may write to it: it approves each source Maurice, one by one. Nobody joins the swarm without being invited.

2. ⑧ You authorise every request. The answer comes back signed.

   • Before sending, Maurice shows you what will leave and its limits. Nothing leaves without your consent.
   • Only the request travels: your conversation stays on your computer.
   • The answer comes back signed, with where it came from and its real state. The conversation can go on over several turns.
   • No silent replay: a request whose outcome is unknown is flagged as such and waits for your decision.

3. ⑨ An unapproved Maurice is refused.

   An unknown Maurice, or one that has not been approved, cannot ask for anything. Its attempt is refused, and you can see it.

4. ⑩ A revoked permission is really revoked.

   When a permission is revoked, the link is cut right away. The next request is refused: no grace period, no forgotten access.

One door to the whole swarm

1. ⑪ From a single Maurice, your AI agent, a messaging app or an API reach the whole swarm, wherever it is allowed.

   You don't have to connect your AI agent to every Maurice. Connect it to yours: Grok Bot, Hark, Muse, Dots, a messaging app or an API go through it to reach the team's Maurice or the company's. Your Maurice relays wherever it is allowed, and only there: every link keeps its own permissions, and the answer comes all the way back to the agent.