Review only the calls your AI wasn't sure about.
Only escalations your AI systems flagged as needing a human land here. Respond with Approve, Reject, or optional feedback, and the AI receives your decision and continues.
Screens
What it does
- 01One inbox for everything
- Escalations from any AI system in one place, colour-coded by level so priority is obvious at a glance.
- 02Real-time updates
- New escalations appear instantly over SSE — no reload needed.
- 03Approve, reject, feedback
- Open an item, read the detail, respond. Optional feedback helps the AI make a better call next time.
- 04Resolved once, for good
- A resolved item can't be resolved again — the record stays consistent for auditing.
- 05Dashboard
- Pending count, breakdown by level, and average resolution time at a glance.
- 06Usable from any AI system
- comet-taskAI, ai-scheduler, your own scripts — anything that can POST over HTTP can connect.
How it works
Launch and issue a key
Issue an API key in Settings and copy the API endpoint shown there.
Configure your AI system
Give the URL and key to whatever sends escalations — they arrive automatically when a decision is needed.
Review and respond
Open an item from the inbox and Approve or Reject. The AI receives the result and continues.
Good to know
- Three levels — Interrupt (paused, waiting), Urgent (attention soon), Emergency stop (outside agreed scope; review first).
- A duplicate approve/reject from another client is rejected with 409.
- Resolution timestamps are server-assigned, keeping aggregates accurate.