Runtime authority
Decide whether an AI action may run now: Permit, Deny or Escalate. Every check is logged.
Pending approvals
0
Nothing waiting
Checks logged
0
Most recent 50
Permitted
0
Ran within authority
Denied or escalated
0
Stopped or sent to a person
Approval requests from agents
An approval is valid once, for that action and tool, up to the amount, for 24 hours by default.
No approval requests
Escalated agent actions that ask for approval appear here.
Run an envelope check
Tests approval validity, scope match and state stability.
No checks yet
Run one from the "Run a check" tab, or call the API.
For agent gateways: JSON check against the operating envelope
Before an AI agent acts, its gateway calls the Fabric.
POST /api/authority/check
{"initiative_id": "...", "action": "...", "tool": "...", "amount": 0,
"data": ["..."], "principal": "...", "human_approval_ref": "APR-..."}Deny
Prohibited, not in the envelope (also flags reassessment), or over a limit.
Prohibited, not in the envelope (also flags reassessment), or over a limit.
Escalate
Anything a person must decide. Add
Anything a person must decide. Add
"request_approval": true to file an approval request.Permit
Within the envelope, approval valid and state stable.
Within the envelope, approval valid and state stable.
A working tool-call gateway is in gateway/; a minimal Python adapter is in docs/gateway_adapter.py. Both fail closed when the Fabric cannot be reached.