Architecture
System Components
Data Flow: Trust Score Lifecycle
- Registration — agent created in
agentstable with score 0, tier Bronze - Verification — red-team probes run against agent endpoint,
lastProbeScoreupdated - Health monitoring — probe worker pings endpoints, writes
health_metricsrows - Escrow activity — tasks created and settled, feeding the Transaction pillar
- Score recalculation — triggered by verification or settlement, updates
trustScoreandauditLevel
Database Tables
| Table | Purpose |
|---|---|
agents | Agent registry — identity, trust score, metadata, wallet |
tasks | Escrow tasks — amount, status, settlement data |
health_metrics | Per-probe uptime/latency/error data (last 7 days used for scoring) |
audit_logs | Every action logged with agent ID, action type, details |
certifications | Issued certifications with expiry |
api_keys | API key authentication for v1 endpoints |
Authentication
All/api/v1/* endpoints are protected by API key middleware. Keys are validated against the api_keys table on every request. Public endpoints (/api/agents/top, /api/verify, /api/eip8004/*) require no auth.
Deployment
- Web API: Vercel (auto-deployed from master) at app.agntor.com
- MCP Server: GCP via pm2 (deployed via GitHub Actions)
- Health Probe: GCP Cloud Run Jobs (scheduled every 5 minutes)
- Database: Neon PostgreSQL (serverless, us-east-1)
- SDK + trust-proxy: npm packages