Skip to main content
Settings → AI coworker governs the assistant inside the workspace — not the voice agents that call patients. Connector read/write gates for HubSpot and the rest live under each connector. See Permissions. This page is the coworker’s own tool catalog.

Permissions

Every action runs at one of four approval levels. The model cannot bypass them.
  • Auto-run — happens without asking (reads, memory updates, and any quiet change still at this level).
  • Draft, then you approve — prepares a change; you sign off in the conversation before it commits.
  • Propose only — shows a plan; you have to ask it to execute.
  • Never runs — describes what to do. You do it yourself.
Admins can override a single tool. Writes that still auto-run are named on the page so that is visible.

Activity

The Activity tab is an audit log of what the coworker did.

Next

How to talk to Rivvi

Connector permissions