Observability
Meridian Blue exposes a Prometheus scrape target, a health endpoint, and structured Conversation rows for every request. Wire them into your existing dashboards — no Meridian-specific stack required.
Health endpoint
GET /health is unauthenticated and returns liveness + database state. Use it as a Kubernetes liveness/readiness probe.
{
"status": "Meridian Blue Server Online",
"db": "connected",
"ts": "2026-04-29T13:34:42.123Z"
}Prometheus metrics
GET /metrics returns a standard Prometheus exposition (text/plain version 0.0.4). Add it to your Prometheus / Grafana Agent / Datadog Agent / Victoria Metrics scrape config.
scrape_configs:
- job_name: meridian-blue
scheme: https
static_configs:
- targets: ["api.meridianblue.ai"]
metrics_path: /metricsAvailable metrics
| Metric | Type | Labels |
|---|---|---|
meridian_requests_total | counter | model, provider, status |
meridian_request_errors_total | counter | model, provider, error_code |
meridian_fallbacks_total | counter | model, attempts |
meridian_request_latency_ms | histogram | model, provider |
meridian_tokens_total | counter | direction (input/output) |
meridian_credits_consumed_total | counter | tenant |
meridian_cost_usd_total | counter | provider, model |
meridian_active_connections | gauge | — |
meridian_provider_attempts_total | counter | provider, model, status |
meridian_risk_classifications_total | counter | risk_level |
Conversation log
Every successful request writes a Conversation row containing input/output content, token counts, total cost (in credits and USD), the provider attempt history, latency, and (when applicable) the coupon ID and discount amount. Query it via the dashboard's Conversations page or the GET /api/v1/developer/usage endpoint.
Audit log
Every authenticated non-GET request and every auth event (login / register / oauth / failed login) writes an ActivityLog row. Query via GET /api/v1/audit-logs with filters for userId, action, and time range. See Audit vault.
Access control
- Conversations — users see their own; admins / owners see the whole tenant; auditors are read-only.
- Audit log — admins, owners, and auditors can read.
reviewercan read entries scoped to the human-review queue. - Metrics —
/metricsis unauthenticated by default. Put it behind your scrape proxy if you want to gate it. - Health — always public.