Whole-home local-first AI platform
Can a home assistant understand context instead of merely following routines?
ANIMA HA is building a whole-home AI and automation platform whose assistant understands household context, investigates meaningful events, and chooses whether to act, speak, notify, research, or do nothing within explicit safety and privacy boundaries.
Blog entries on recent progress from the researchers
No entries yet. New project notes will appear here.
Anima should know the household, notice selected meaningful events, investigate when needed, and choose an appropriate response rather than act as a pile of fixed smart-home rules. Its core must remain local-first, permission-aware, auditable, and safe when context is uncertain.
Phase 12 is now accepted: one configured sample household can be operated through the responsive interface, with household truth, policy, confirmation, tasks, calendar, audit, and layout configuration flowing through the shared core rather than a customer-specific fork. The acceptance includes API and browser evidence, but it preserves several recovery, pagination, rejection-label, and fencing defects for later work.
Owner-directed alert, drawer, and presence changes are reported, but the published source and current image have not independently established them. Vendor delivery, live wake checks, source-bound sensor behavior, and the three-day household trial remain open, so final goal completion is still unclaimed.
ANIMA HA is an authorized working-prototype program, not yet a production home assistant or unrestricted autonomous system.
Read the source record in Notion ↗