Skip to content

Safe Entity Exposure

Jump to bottom
itsreverence edited this page Jun 3, 2026 · 1 revision

Safe Entity Exposure

Codex Assist should not bypass Home Assistant safety controls. It routes control through Home Assistant's normal Assist LLM API, so the practical boundary is your Assist exposed-entity list.

The basic rule

Only expose entities you actually want an AI conversation agent to see or control.

Good first test entities:

  • one desk lamp
  • one common-room light
  • read-only sensors
  • harmless helpers or test switches

Think carefully before exposing:

  • locks
  • garage doors
  • alarms
  • water shutoff valves
  • covers or blinds
  • security controls
  • cameras
  • anything that could surprise family members

Suggested first setup

  1. Expose a single harmless light.
  2. Ask Codex Assist what it can see.
  3. Test turning that light on and off.
  4. Add more entities only after the basic flow behaves as expected.

If control does not work

Check these before assuming the model failed:

  • Is the target entity exposed to Assist?
  • Is your Assist pipeline using Codex Assist?
  • Is the entity name or alias clear enough?
  • Does the device work from the normal Home Assistant UI?

Safety stance

For the canonical security statement, see SECURITY.md.

Clone this wiki locally