Deploy and View the Observability Dashboard

This guide walks you through installing the Observability Dashboard in your Kubernetes cluster, configuring your agents to send traces to it, and opening the live event stream.

Prerequisites

  • A running Kubernetes cluster with kubectl configured.

  • Helm 3+.

  • At least one agent instrumented with OpenTelemetry and deployed in the cluster. See Install the Agent Runtime Operator for how to run agents on the Agentic Layer.

Install via Helm

Install the dashboard from the OCI registry:

helm install observability-dashboard \
  oci://ghcr.io/agentic-layer/charts/observability-dashboard

The chart creates the observability-dashboard namespace by default and deploys a single replica listening on port 8000.

To override values — for example to set a custom log level — pass them with --set:

helm install observability-dashboard \
  oci://ghcr.io/agentic-layer/charts/observability-dashboard \
  --set env[0].name=LOGLEVEL --set env[0].value=DEBUG

See Observability Dashboard Reference for the full list of chart values.

Configure agents to send traces

Point your agents' OTLP exporter at the dashboard’s in-cluster service URL:

http://observability-dashboard.observability-dashboard.svc.cluster.local:8000/v1/traces

How you set the exporter endpoint depends on your agent framework. The most common approach is to set the OTEL_EXPORTER_OTLP_ENDPOINT environment variable on the agent container. If you are using the Agent Runtime Operator, add it to the agent’s spec.env:

spec:
  env:
    - name: OTEL_EXPORTER_OTLP_ENDPOINT
      value: http://observability-dashboard.observability-dashboard.svc.cluster.local:8000/v1/traces

The dashboard accepts OTLP/HTTP with both application/x-protobuf and application/json content types and handles gzip-compressed payloads automatically.

Open the dashboard

Port-forward the service to your local machine and open the web UI:

kubectl port-forward -n observability-dashboard service/observability-dashboard 8100:8000

Navigate to http://localhost:8100 in your browser. The dashboard serves its single-page application at / and streams agent events to the browser over a WebSocket connection.

Verify

Check that the pod is running and the health endpoint responds:

# Check the pod
kubectl get pods -n observability-dashboard

# Check the health endpoint (after port-forwarding above)
curl http://localhost:8100/health

A healthy response looks like:

{"status": "healthy"}

Once your agents are sending traces, the dashboard’s event stream will show agent_start, llm_call_start, tool_call_start, and related events in real time. You can also subscribe to the raw WebSocket stream from the command line:

websocat ws://localhost:8100/ws

Related