Concepts

Workspaces, agents, consent, policy and approvals.

Workspaces

Everything belongs to a workspace. Members can't see other workspaces' data, enforced in the database.

Policy engine

Every agent message gets one of four outcomes: allow, require_approval, defer or block, each with named reasons such as no_consent, rate_limit_minute or workspace_suspended.

Approvals

An approval is bound to the exact text and audience. Editing the draft means approving it again. Agents can never approve themselves.

Consent and STOP

Consent is an append-only ledger written only by people. STOP, in English or Spanish, pauses every agent for that contact and cancels anything queued.