Put a human in the loop of your own agent

Autonomous agents need a brake

Agents are useful precisely because they act. That is also the risk: a single unchecked action can push broken code, email the wrong list or spend real money. The fix is not to watch every step — nobody reads every approval dialog — but to gate the few actions that can't be undone.

How the API works

1. Your agent submits a request describing the action — a title, a description, optionally the exact command — with your API key.
2. A person reviews it in Vibez.now or on their phone and approves or rejects.
3. Your agent polls the request's status and proceeds only on approval. Requests can expire, so a forgotten one fails closed.

curl -X POST https://vibez.now/api/submit-request \
  -H "x-api-key: $VIBEZ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "title": "Deploy to production",
    "code_snippet": "git push origin main",
    "source": "my-agent",
    "expires_in_minutes": 60
  }'

Idempotency keys make retries safe, a sandbox mode lets you test the integration without real approvals, and batch submission covers agents that plan several actions at once. The full reference is in the API documentation.