Operator configuration

Configure the OpenLIT Operator deployment with Helm values, and define AutoInstrumentation custom resources to control how workloads are instrumented.

Operator configuration is split into two areas:

1. Helm values — control the operator deployment itself (replicas, resources, log level, images)
2. AutoInstrumentation CRD — control which pods are instrumented and how

Operator Helm values

Pass values at install or upgrade time with --set flags or a values.yaml file.

Deployment

deployment.replicas · integer · default: "1"

Number of operator replicas. For high availability, set to 3 and configure podAntiAffinity.

deployment.strategy.type · string · default: "RollingUpdate"

Kubernetes deployment strategy. Use Recreate if you need to avoid running two operator versions simultaneously.

deployment.priorityClassName · string · default: ""

Priority class for the operator pod. Set to system-cluster-critical for production environments.

Resource limits

resources.requests.cpu · string · default: "100m"

CPU request for the operator pod.

resources.requests.memory · string · default: "128Mi"

Memory request for the operator pod.

resources.limits.cpu · string · default: "500m"

CPU limit for the operator pod.

resources.limits.memory · string · default: "512Mi"

Memory limit for the operator pod.

Example — high availability deployment:

deployment:
  replicas: 3
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxUnavailable: 1
      maxSurge: 1
  priorityClassName: system-cluster-critical
  affinity:
    podAntiAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
      - weight: 100
        podAffinityTerm:
          labelSelector:
            matchLabels:
              app.kubernetes.io/name: openlit-operator
          topologyKey: kubernetes.io/hostname

resources:
  requests:
    cpu: 200m
    memory: 256Mi
  limits:
    cpu: 500m
    memory: 512Mi

Observability

observability.logLevel · string · default: "info"

Operator log level. Accepts debug, info, warn, or error.

observability.selfMonitoringEnabled · boolean · default: "false"

Emit the operator’s own telemetry via OpenTelemetry. Requires observability.otel.endpoint.

observability.otel.endpoint · string · default: ""

OTLP endpoint to send operator telemetry to, for example http://openlit:4318.

Example:

observability:
  logLevel: info
  selfMonitoringEnabled: true
  otel:
    endpoint: "http://openlit:4318"
    headers: "Authorization=Bearer my-secret-token"

Provider images

The operator uses pre-built init container images for each instrumentation provider. By default these inherit the operator’s own image tag.

providerImages.openlit.repository · string · default: "ghcr.io/openlit/openlit-ai-instrumentation"

Image repository for the OpenLIT instrumentation provider.

providerImages.openllmetry.repository · string · default: "ghcr.io/openlit/openllmetry-ai-instrumentation"

Image repository for the OpenLLMetry instrumentation provider.

providerImages.openinference.repository · string

Image repository for the OpenInference instrumentation provider.

Example — using a private registry:

providerImages:
  openlit:
    repository: my-registry.example.com/openlit-ai-instrumentation
    tag: "v1.0.0"
  openllmetry:
    repository: my-registry.example.com/openllmetry-ai-instrumentation
  openinference:
    repository: my-registry.example.com/openinference-ai-instrumentation

Instrumentation defaults

instrumentation.defaultProvider · string · default: "openlit"

Default instrumentation provider used when none is specified in the AutoInstrumentation CR. Accepts openlit, openinference, openllmetry, or custom.

instrumentation.defaultVersion · string · default: "latest"

Default provider image tag.

instrumentation.defaultImagePullPolicy · string · default: "IfNotPresent"

Image pull policy for init containers.

---

AutoInstrumentation CRD

The AutoInstrumentation custom resource defines which pods to instrument, how to instrument them, and where to send telemetry.

Full spec reference

apiVersion: openlit.io/v1alpha1
kind: AutoInstrumentation
metadata:
  name: my-instrumentation
  namespace: default
spec:
  # Pod selection — required
  selector:
    matchLabels:
      openlit.io/instrument: "true"
    matchExpressions:
    - key: "environment"
      operator: In
      values: ["production", "staging"]

  # Pods to skip even when they match the selector
  ignore:
    matchLabels:
      openlit.io/skip: "true"

  # Python instrumentation settings
  python:
    instrumentation:
      enabled: true
      provider: "openlit"        # openlit | openinference | openllmetry | custom
      version: "latest"
      imagePullPolicy: "IfNotPresent"
      customPackages: "langchain>=0.1.0,chromadb>=0.4.0"
      env:
      - name: OPENLIT_ENVIRONMENT
        value: "production"
      - name: OPENLIT_API_KEY
        valueFrom:
          secretKeyRef:
            name: openlit-secret
            key: api-key

  # OTLP endpoint — required
  otlp:
    endpoint: "http://openlit.openlit.svc.cluster.local:4318"
    headers: "authorization=Bearer token123"
    timeout: 30

  # Resource attributes added to all telemetry
  resource:
    environment: "production"
    serviceName: "ai-chat-service"
    serviceNamespace: "ai-platform"
    serviceVersion: "v2.1.0"

Pod selection

spec.selector · LabelSelector · required

Pods matching this selector in the same namespace will be instrumented. Supports matchLabels (exact match) and matchExpressions (operators: In, NotIn, Exists, DoesNotExist).

spec.ignore · LabelSelector

Pods matching this selector are skipped, even if they match spec.selector. Use to exclude system components or debug pods.

AutoInstrumentation resources only affect pods in the same namespace. To instrument pods across namespaces, create a separate AutoInstrumentation resource in each namespace.

Python instrumentation

spec.python.instrumentation.enabled · boolean · default: "true"

Enable or disable Python instrumentation.

spec.python.instrumentation.provider · string · default: "openlit"

Instrumentation provider. Accepted values:

Value	Description	Init image
openlit	Full AI observability with cost tracking	ghcr.io/openlit/openlit-ai-instrumentation
openinference	OpenInference conventions, OTel-standard	ghcr.io/openlit/openinference-ai-instrumentation
openllmetry	LLM-focused OpenTelemetry extensions	ghcr.io/openlit/openllmetry-ai-instrumentation
custom	Your own init container image	Set customInitImage

spec.python.instrumentation.version · string · default: "latest"

Provider image tag to use.

spec.python.instrumentation.customPackages · string · default: ""

Comma-separated list of additional pip packages to install in the init container, e.g. langchain>=0.1.0,chromadb>=0.4.0.

spec.python.instrumentation.customInitImage · string · default: ""

Full image reference for a custom init container. Required when provider is custom.

OTLP configuration

spec.otlp.endpoint · string · required

OTLP HTTP endpoint. Must be a valid HTTP or HTTPS URL.

Format	Example	Use case
In-cluster service	http://openlit:4318	Same namespace
Fully qualified DNS	http://openlit.openlit.svc.cluster.local:4318	Cross-namespace
External HTTPS	https://traces.example.com:4318	Cloud / external

spec.otlp.headers · string · default: ""

HTTP headers as comma-separated key=value pairs, e.g. authorization=Bearer token123,x-api-key=secret. Used for authentication with external backends.

spec.otlp.timeout · integer · default: "30"

Request timeout in seconds. Accepted range: 1–300.

Resource attributes

Resource attributes are attached to every span and metric as OpenTelemetry resource attributes. The operator also automatically adds k8s.pod.name, k8s.namespace.name, k8s.deployment.name, k8s.node.name, and related Kubernetes attributes.

spec.resource.environment · string · default: ""

Maps to deployment.environment. Example: production.

spec.resource.serviceName · string · default: ""

Maps to service.name. Overrides the auto-detected service name.

spec.resource.serviceNamespace · string · default: ""

Maps to service.namespace.

spec.resource.serviceVersion · string · default: ""

Maps to service.version. Example: v2.1.0.