Home/Docs/System PRD

Personal AI workspace: practical design brief

Updated October 30, 2026. Educational guidance; use current official documentation and your installed version.

This replaces the older aspirational system specification. It describes responsibilities and acceptance criteria, not a claim that every integration or automation is deployed.

Goal

Help a human prepare, decide, and follow through with less repetitive coordination. Optimize for useful verified outcomes, not the number of agent names or background runs.

Components

  • OpenClaw: conversational interface, connected tools, memory, sessions, and scheduled jobs.
  • Paperclip, optional: durable specialist tasks, ownership, execution history, and handoffs.
  • Mission Control, optional: private operational dashboard and controls, not a second competing agent scheduler.
  • Source applications: authoritative records for mail, courses, documents, calendars, and task deliverables.

Minimal acceptance criteria

  1. One known read succeeds from the intended account.
  2. One bounded task produces a useful artifact, with source references.
  3. An unchanged scheduled check uses no model and stays quiet.
  4. A relevant change creates at most one tracked handoff.
  5. Access failure is unknown, not zero; retry is bounded.
  6. A failed/ambiguous save is reconciled before another save.
  7. No sending, publication, or grade change occurs outside its authorization.
  8. The owner can inspect state, stop the schedule, and recover from backup.

Growth path

First: one assistant and one tool. Next: one deterministic monitor and one reliable specialist. Later: independent review, multiple portfolios, and a dashboard if coordination justifies them.

Configured, queued, running, drafted, delivered, reviewed, and completed are different states. Keep them visible. Do not make a dashboard's green status stand in for an end-to-end workflow test.

See architecture, Paperclip, and the dated setup overview.