Status & SLA

Service levels are measured and published, not just promised. Data
comes from synthetic probes every 5 minutes (database round-trip plus a
real hybrid-search query) and from every real request across the MCP,
REST, and crawler-landing surfaces.

Targets

Endpoints

• Human: /status —
  SLO verdicts (met / missed / measuring), per-surface traffic with
  p50/p95/p99, and how the platform scales.
• Machine: GET /api/status?days=<1–90> — the same report as JSON,
  CORS-open, for your own monitoring:

{
  "windowDays": 7,
  "uptime": { "checks": 2016, "okPct": 99.95, "dbLatencyP50Ms": 4, … },
  "surfaces": [
    { "surface": "MCP (agents)", "calls": 51234, "errorRatePct": 0.1, "p50Ms": 38, "p95Ms": 210, "p99Ms": 420 }, …
  ],
  "slo": {
    "availability": { "targetPct": 99.9, "measuredPct": 99.95, "met": true },
    "searchP95":    { "targetMs": 300, "measuredMs": 41, "met": true },
    "errorRate":    { "targetPct": 1, "measuredPct": 0.12, "met": true }
  }
}

• Liveness: GET /api/health — 200 {ok: true, dbLatencyMs} or 503;
  point your uptime monitor here.

Contractual SLAs with credits come with enterprise agreements; the
measured numbers above are what we'll sign against.