Features · Approvals
Risky waits.
A person decides, and it is written down.
Suspended, then decided.
When an agent wants to make a risky call, the call is held in the agent and an approval appears in the inbox with what, which agent and since when. A person approves or denies. The hub records the decision and tells the agent with a signed call, and the agent carries on or tells the model it was denied.
- A decision is recorded
- Only a member of the workspace decides
- Your AI can see the queue but not answer it
Honest labels
What this guards.
- LiveRisky tool calls
Webhooks and connector tools not marked read-only.
- Not offeredAnything you mark safe
If you mark a write tool as read-only, the gate lets it through. Mark carefully.
Questions
About approvals.
What counts as risky?
Every connector tool unless you marked it read-only, a webhook always, and any tool the rule does not know. Web fetch and web search are safe.
Who can approve?
A person of the workspace, in the app. The AI door has no approve action.
Can my AI approve things?
No. There is no approve action, so an AI cannot approve its own request.