First tagged release of Janus — an AI safety gatekeeper between any LLM and Home Assistant: the model proposes, deterministic code decides. Dangerous operations always confirm; ambiguity asks; vague intent becomes a suggestion, never a silent action.
- HA conversation agent (install via HACS), CLI REPL, and a headless
/v1service - Local (Ollama) or cloud (Anthropic) LLM backends
- Registry curation, disambiguation, intent inference, decision audit, dangerous-op PIN
- One-command full stack (
deploy/compose) and a HAOS add-on (deploy/addon)