Hi everyone,
We built Spatial OS around one constraint: a language model should not send opaque commands directly to physical devices.
The current Home Assistant flow is:
- Discover rooms and devices through Home Assistant OAuth.
- Ground the requested outcome in room, device, capability and relationship context.
- Compile it into typed, inspectable actions.
- Run a shadow simulation before anything physical happens.
- Require explicit approval for that single execution.
- Verify observed device state afterward and attempt rollback for reversible actions when needed.
The complete Virtual Home Lab works in the browser without hardware, so the policy, simulation, approval and verification flow can be tested before connecting a real home.
Demo: Spatial OS | Deterministic home intelligence
We are looking for five Home Assistant users who can give blunt feedback on:
- whether this safety model is genuinely useful or just extra ceremony;
- which setup breaks the one-minute connection promise;
- whether room and device grounding matches real homes;
- what evidence you need before approving a live run;
- which actions should never be delegated.
This is a founding beta, not a claim of universal device compatibility. Home Assistant is the only live connector today, critical actions remain blocked, and natural language never talks to hardware directly.
— Spatial OS Team