Which unsupervised action hurts first?
Start with the action that posts, pays, or changes data. Protect that action first.
Set clear limits before assistants act.
Stop unapproved posts, payments, and data changes. The connected tool checks a current permission receipt and refuses work outside the approved limits.
Clear limits before assistants act
Who approves the agent?
A person owns the assistant and sets its limits. The connected tool enforces what the receipt allows.
What can reviewers see?
Reviewers can open the permission receipt and see the allowed actions, spending limit, and expiration date.
Set limits before assistants act in the NEUS productHow it works
Set owner, limits, check, expand
- 1Pick the painful tool pathChoose one production action where unsupervised failure is expensive.
- 2Check permission on that pathBlock the call until a current agent permission receipt exists.
- 3Show refuse-on-expireRevoke or expire the receipt and confirm the host refuses the next attempt.
- 4Expand to the next surfaceUse the same owner and limits in assistant tools, developer tools, and production.
Agent permission
Approved
Agentops-assistant
OwnerAcme Ops
Allowedread, draft, notify
Deniedtransfer, delete
Spend cap$25.00
Expires7 days
Connected tools confirm limits before a protected action runs.
Integration options
Use the path closest to your first protected decision.
- Hosted check
- Start without building a custom screen
- JavaScript
- Add the check to a product backend
- HTTP
- Check a receipt from any service
- React
- Drop the flow into an existing interface
- Assistant setup
- Bring trust checks into developer tools
What stays true
Clear checks at the moment they matter.
- Clear approval before a sensitive action
- Out-of-scope work is refused before it runs
- Set detailed limits when your first protected action is ready
Questions about assistant limits
How approval works before an assistant takes a sensitive action.