Core terms

• Operator — the always-on agent that watches your linked accounts and drafts changes. It never applies a change itself.
• Draft — a proposed change shown as a before → after diff. Inert: it touches nothing in your ad account until you approve it.
• Approval queue — the screen where you review drafts and run the two-step Approve & push live → Confirm control.
• Autonomy loop — the continuous propose-only cycle that finds money and stages drafts. It does not auto-apply in V1.
• Audit log — the append-only, KMS-signed record of every approved action: who approved it, what changed, and when. Source of one-click rollback.
• Rollback — reverting an approved change to its prior state in one click, using the value the audit log captured.
• Daily envelope — the 24-hour limit on how much change the loop can stage, so a bad day cannot compound.
• Circuit breaker — the safeguard that halts the loop when signals look wrong, before it drafts more.
• Read — pulling account, structure, and performance data. Ungated on every plan; reading never spends.
• propose_change — the single MCP write tool that creates a draft. It cannot push a change live.
• mcp.read — the OAuth scope that lets your AI client read account data.
• mcp.draft — the OAuth scope that lets your AI client create drafts. There is no mcp.approve scope.
• Connected account — an ad account linked over the platform OAuth. Links start read-only.
• Workspace — an Agency-tier container that isolates one client account from another, each with its own audit log.