This page details the operational mechanics of the Fraink agent within the fraink-platform.

Core Operational Loop

The agent operates in a 5-step loop constrained by user permissions:

1. Ask: User interacts via chat.
2. Read: Agent reviews folder context (CRM, catalog, support).
3. Propose: Agent drafts a change with a preview/diff.
4. Apply: Executed strictly within the human connector's active permissions.
5. Record: Action is logged under the human's identity.

The Authority Gate

The agent has no independent identity or elevated privileges. It acts solely as the user who connected it. If the user can make the change in the management UI, the agent can; otherwise, it blocks. Permissions map directly to the existing organization chart (e.g., franchisee vs. franchisor scope).

Autonomous Drafting

Without explicit prompts, the agent reads records, maintains threads contextually alongside the work, and drafts subsequent moves (notes, status changes, page edits) as previews.