Context Capture Boundary

Personal assistants get more useful as they capture more context, but every additional input stream (screen, transcripts, inbox, chat) adds a governance obligation. The framework says to draw the capture boundary deliberately: decide which client data enters the assistant's memory before deployment, not after. Agencies that skip this step end up with a searchable archive they cannot cleanly separate per client, which turns a productivity tool into a liability during audits or offboarding. A concrete example: MyHandler builds an encrypted on-device database from screen content, meeting transcripts, files, and messages across email, Slack, Teams, SMS, Telegram, and Discord, so an agency running it across three retainers must decide upfront whether one shared memory is acceptable or whether each client needs an isolated instance. The same logic applies to PHNTM One, which processes voice, documents, and personal data locally with no cloud connectivity, trading breadth for a narrower, easier-to-govern footprint.