ObservaKit

Commercial APM and observability suites bill in hundreds of dollars per month, and most teams use a fraction of the surface area.

BYOK and data residency are usually bolt-ons, when they should be the default.

Recovery actions sit behind dashboards — humans still have to read graphs at 3am to restart a service.

OpenTelemetry collectors feed a NATS-based event bus; Grafana OSS handles visualization.

Local Ollama sidecar runs the AI layer — summaries, anomaly hypotheses, root-cause sketches — without sending telemetry to a third party.

Autonomous recovery agents act on a narrow allow-list of remediations (restart, scale, failover) with full audit trail.

BYOK by default. Operators bring their own credentials; ObservaKit never holds long-lived secrets.

Cost ceiling is a feature, not a side effect — the whole stack is sized to run on a small VPS.

Every automated action is reversible, logged, and gated by policy.