Personal Assistant Operating Model

My home personal assistant is a control layer for projects, notes, handoffs, automation ideas, and technical work. It is not one giant agent with unlimited authority. It is closer to an operating model: context, repo boundaries, repeatable workflows, evidence capture, and clear approval gates.

I think of it as a practical assistant with memory, but also with rules. It should know enough to be useful, stay scoped enough to be safe, and keep enough evidence that future work can be resumed without guessing.

The work-side assistant pattern is deliberately separate. I describe it here only at the boundary level: it should operate inside its own work context, use work-approved communication channels, avoid pulling private home context into work decisions, and produce handoffs that are clear enough for a human or another assistant to review.

The interesting design problem is the space between them. Personal and work assistants can cooperate through sanitized briefs, explicit prompts, and reviewed outputs, but they should not casually share everything. Separation is a feature, not a limitation.

That is where the security architecture matters: identity, context, tool access, data movement, approval, and auditability all need to be visible enough to reason about.