A useful assistant should not need the same background every time you return to a task. If you allow it to retain a preference, a responsibility, or the reason for a decision, that context should remain available when you need help again.

The aim here is to keep that knowledge under its owner’s control and connect it to useful work. Models and providers will change. The accumulated context should remain useful.

Household help is more than a calendar.

Calendar intake is an early implementation because notices and events are concrete things to work with. Household OS has a broader purpose: retain permitted household knowledge and connect plans with preparation and follow-through.

Consider a synthetic school notice. Recording the trip date is one step. Remembering that a packed lunch is needed, connecting preparation to the event, and asking before submitting a form are different steps with different requirements.

Assisted intake and public calendar-delivery code provide early building blocks. Event-linked preparation is being built. Consequential actions and broader proactive assistance remain directions. A saved receipt does not prove that a service is running continuously.

The right result may not be a conversation.

Sometimes you need an answer. Sometimes a calendar entry, task, reminder, or approval request is more useful. When there is nothing helpful to add, the system should stay quiet.

That also matters in business. Sovereign Vault explores a separate application of the foundations, with its own data and permission rules. Household and business use should each be designed around the people doing the work.

Why start with a protocol?

Sovereign Memory Protocol is the original foundation: shared rules for keeping authority over information and preserving where it came from and how it changed. Its public draft remains under development. Public access does not establish implementation conformance.

Core provides a PostgreSQL reference implementation. User MCP explores bounded access. Applications choose appropriate implementations and policies. This separation leaves room to adapt existing systems instead of tying the entire design to one tool.

What future notes should establish.

Each note should explain a problem, what was tried, what the evidence supports, and what is still missing. Core’s v0.3-alpha release, for example, publishes bounded restore evidence rather than a promise of universal production readiness.

A standalone interface outside a frontier-model app remains a direction. No finished self-serve household product or full automation is offered here.

Prepared with AI assistance from Jesse’s stated aims, a sanitized project review, and public repository documentation. Examples are synthetic.