# observability.md

_Last updated: 2026-08-15_
_Version: 1.0.0 - canonical: https://encrata.com/ai/observability.md_

How Encrata watches its own health, and where you should look when something seems wrong.

## For users and agents

- **Status page**: https://status.encrata.com - uptime, incidents, maintenance windows. Check here first.
- **Request IDs**: every API response carries a `request_id`. Include it in support requests (support@encrata.com) for fast tracing.
- **Rate-limit signals**: `429` responses include `Retry-After`. Treat them as expected and retryable, with backoff.

## What we run

- **Error tracking**: Sentry on the web frontends with release tagging, so production stack traces map to source.
- **Client telemetry**: Grafana Faro (first-party collector, same-origin) for web vitals and front-end errors.
- **Analytics**: privacy-conscious product analytics (PostHog); no third-party ad trackers.
- **API metrics**: latency, error rate, credit spend and cache hit rate per endpoint, with alerting on error-rate and latency budgets.
- **Uptime probes**: external checks against production endpoints feed the status page.

## Logging boundaries

Usage logs (endpoint, status, timing, credits) are kept 90 days. Lookup response bodies, passwords and API keys are never logged. See https://encrata.com/governance.