Skip to main content
The Innate agent is our embodied agent running in the cloud. Think of it as the LLM OS running your agents and skills: it reasons over live perception, chooses the right skills, and controls the robot autonomously from your instructions.
Naming: “the Innate agent” is the cloud AI system itself. The “agents” you write (see Agents) are Python definitions — a prompt, skills, and inputs — that the Innate agent runs.
The Innate agent is a multi-model system that behaves like one assistant with memory and planning. A key capability is spatial RAG (Retrieval-Augmented Generation over spatial memories): what MARS sees gets stored as memories tied to places, so the agent can retrieve past observations (objects, places, scenes), answer questions about them, and navigate back to where they were seen. In practice, this means you can ask things like “where did you last see my keys?” and the Innate agent can use its memory to point to a location or return there. The Innate Controller app surfaces these memories on your phone, so you can inspect what was remembered and where.

​
Connecting to the Innate agent

The Innate agent is accessible for free to all users of Innate robots for 300 cumulative hours — and probably more if you ask us. The Innate agent runs in the cloud at wss://agent-v1.innate.bot. Your MARS robot connects to it automatically over WebSockets using Innate’s protocol — the connection is handled by the brain_client_node in Innate OS, so you don’t need to configure anything manually. (If you ever need to point at a different endpoint, set BRAIN_WEBSOCKET_URI in ~/innate-os/.env.)

​
Authentication

Each MARS robot is given a unique token that authenticates it with the Innate agent. It is stored as INNATE_SERVICE_KEY in ~/innate-os/.env on your robot — to check it, SSH in and run cat ~/innate-os/.env. If you’ve accidentally deleted your token or need to retrieve it, reach out to us on Discord or via email and we’ll provide it.