Skip to content

Log 008: Household Gods: Permission Layers in Domestic AI

Domestic-scale orchestration where the household is a contested multi-agent system.

  • Mode: Thinking Space
  • Status: Working Note
  • Date: April 2026
  • Scope: domestic orchestration and permission layers
  • Depends on: Chord vs Arpeggio, Epistemic Firewalls, Human Vital Systems

Problem

Most assistant architectures treat "the household" as one user. This is wrong on contact. Homes are plural polities: unequal power, uneven memory, conflicting consent.

Proposal

Model home assistants as permission spirits rather than omniscient servants: - memory steward (what is remembered, what expires), - boundary steward (who may ask what, when), - escalation steward (what leaves the house).

The key design principle is non-unitary consent. "Family preference" is often a coercive aggregate.

Failure cases to design for

  1. Grief lock-in: dead relatives become permanently active through never-expiring memory loops.
  2. Domestic capture: dominant member weaponizes assistant access policies.
  3. Care/control inversion: safety checks become surveillance theater.

Architectural consequences

  • Require per-member veto lanes and temporal consent (revocable later).
  • Default to partial forgetting, not infinite retention.
  • Use local-first logs with explicit export rituals, not silent cloud drift.

Open design question

What is the minimum review bandwidth for a household to remain governable without externalizing every intimacy to platform arbitration?