Skip to main content
Sentinel watches your systems. It catches problems before they become failures, logs every error as a lesson, and attempts self-healing before escalating. It’s the agent that runs in the background so every other agent can run cleanly.

​
What Sentinel monitors

Sentinel health check run
├── API connectivity
│   ├── Ultron platform endpoints
│   ├── Connected integrations (Gmail, HubSpot, Stripe, etc.)
│   └── MCP server availability
├── Auth status
│   ├── OAuth token expiry checks
│   └── API key validity
├── Performance
│   ├── Response time baselines
│   └── Error rate trends
└── Self-checks
    ├── Memory store integrity
    └── Lessons database review

​
Error handling

When Sentinel detects a problem:
  1. Attempt self-healing — retry with backoff, re-auth, clear cache
  2. Log the lesson — store what broke and how to prevent it next time
  3. Alert if unresolved — Slack message or email to the configured recipient
Sentinel never escalates without first attempting a fix.

​
Lessons logging

Every error Sentinel catches becomes a lessons-system entry: 
{
  "type": "feedback",
  "tags": ["lesson", "sentinel", "auth"],
  "text": "LESSON: Gmail OAuth token expired mid-session without warning. PREVENTION: Check token expiry at session start, refresh if within 24 hours of expiry."
}
Next run, Sentinel checks for its own lessons first and applies them before starting any task.

​
Alerting

Sentinel sends alerts via connected channels when critical thresholds are hit:
  • Slack — message to your configured #alerts channel
  • Email — via Gmail or Brevo to your notification address
  • Threshold examples — API error rate > 10%, auth token expired, uptime < 99.5%

​
Self-improvement over time

Sentinel is the primary contributor to the lessons system. Over weeks of operation, the lessons database accumulates a complete picture of every failure mode in your configuration — and the fix for each one. The entire Ultron agent team benefits from what Sentinel learns.