Control Tower
Usage, cost, oversight, release readiness and value for every AI workflow. Anyone can view it; changing settings needs the admin passcode.
Release readiness
Value ledger
Models and the release gate
Model registry
Worker and supervisor must differ. A change needs a passing evaluation on every workflow.
Evaluation
AI usage and cost
Calls per day
Gateway protections
Page assistant
The "Explain this screen" chat. It answers only from the page, quotes are checked in code, and the supervisor model can withhold an answer. Question text is not stored.
Routing and contracts
Hops by agent
Blocked writes
Agents that tried to touch keys outside their contract.
Controls
Kill switches
A paused agent stops the case and sends it to a person.
Platform
Learning loop
When a person changes a decision, the case becomes a test-set candidate. Accepting it adds it to the evaluation set, so the gate stays red until the rule pack or prompts handle it.
AI risk cards
Audit trail
Append-only and hash-chained: editing or deleting any entry breaks verification from that point.