OBSERVE Working beta by Sovix

Datadog-grade visibility without Datadog-grade bills.

OBSERVE is self-hosted, AI-first observability for Kubernetes. One Helm install gives you metrics, logs, traces, eBPF auto-instrumentation, security events, network flows, SLOs, incidents, and an eleven-agent AI swarm for root-cause analysis, inside your own cluster.

Open the live demo A real k3s cluster with the full stack deployed from the chart.
One Helm install Runs in your cluster MCP server for your agents
From signal to root cause

Every hop is instrumented. When one of them goes wrong, the swarm says which.

eBPF traces every request between services, with no code changes. Metrics, logs and traces land in one place, from the same chart. An error rate crosses its burn-rate threshold and an incident opens. Specialised agents read the evidence and name the likely cause.

Incident
T
TraceseBPF L7 spans for HTTP, gRPC, SQL, Redis and DNS, from a DaemonSet.
M
Metrics and logsTime series and log lines from the same install, queried side by side.
!
SLO burn rateThe error budget for the auth service is burning faster than its window allows.
AI
Root cause, with evidenceThe swarm's analysts read the traces, logs and metrics around the change and name the likely cause, citing what they read.

What one install gives you.

Everything below is provisioned by the Helm chart, in the namespace you choose, with no runtime database for alerting.

  • Metrics

    VictoriaMetrics with vmagent, kube-state-metrics and node-exporter.

  • Logs and traces

    ClickHouse with the OpenTelemetry Collector receiving and routing telemetry.

  • eBPF auto-instrumentation

    Beyla L7 spans for HTTP, gRPC, SQL, Redis and DNS, with zero code changes.

  • Network and security

    Retina for eBPF network flows and policies; Tetragon for syscall-level runtime security events.

  • SLOs and incidents

    Burn-rate tracking against VictoriaMetrics; ClickHouse-backed incident management and log fingerprinting.

  • AI swarm RCA

    Eleven specialised agents (metrics analyst, log deep-diver, pathfinder, sentinel, finops, oracle, architect and more).

  • MCP server

    Exposes the same tools to external agents such as Claude, Cursor and Copilot.

  • Alerting and access

    Alertmanager and vmalert, ConfigMap-driven. Project scoping, multi-cluster discovery and RBAC.

The interface.

Two pages from the app. Both were captured with the demo data mode on, which the badge in the top bar shows.

The OBSERVE overview page: a dark dashboard showing nodes, running pods, namespaces, average CPU and memory, a shift briefing and an intelligence feed.
Overview: cluster health, a shift briefing and the intelligence feed.
The OBSERVE service page for auth-service: request rate, success rate, error percentages, an AI insight panel and charts of request rate by method and latency percentiles.
A service page: rates, latency percentiles, errors and an AI insight.
Install

One chart. NodePort, a hostname, or your existing ingress.

The chart carries the data plane and the app. Pin the chart and image versions you want; the live demo runs the same chart on a small k3s cluster.

Open the live demo

See real eBPF data flowing.

The demo cluster is public. OBSERVE is a working beta from Sovix; ask us about running it in your own cluster.

Open the live demo Ask about OBSERVE