Installation

Full installation guide with all configuration options.

4 minute read

This guide covers installation from prerequisites through custom configuration, verification, and uninstalling.

Prerequisites

Requirement	Minimum version
Kubernetes	1.28+
Helm	3.12+
kubectl	1.28+
Container runtime	containerd, CRI-O
Storage	Default StorageClass with dynamic provisioning (for memory + Tempo PVCs)

Required secrets (created in the agents namespace):

At least one LLM API key (Anthropic, OpenAI, or Google) in a Kubernetes Secret

Full install (all components)

The default install deploys the operator, console, memory service, and Tempo:

helm install agentops oci://ghcr.io/samyn92/charts/agentops-platform \
  --namespace agent-system --create-namespace

This creates two namespaces:

Namespace	Contents
`agent-system`	Operator, console, Tempo
`agents`	Memory service, agent pods, AgentTool/AgentResource/Channel workloads

Minimal install (operator only)

If you only need the operator and will run agents without the console, memory, or tracing:

helm install agentops oci://ghcr.io/samyn92/charts/agentops-platform \
  --namespace agent-system --create-namespace \
  --set agentops-console.enabled=false \
  --set memory.enabled=false \
  --set tempo.enabled=false

The operator installs the CRDs and watches for Agent, AgentRun, AgentTool, AgentResource, and Channel resources. Agents will still function — they just won’t have memory integration or a web UI.

Custom configuration

Using a values file

Create a values-override.yaml and pass it at install time:

helm install agentops oci://ghcr.io/samyn92/charts/agentops-platform \
  --namespace agent-system --create-namespace \
  -f values-override.yaml

Model providers

LLM API keys are configured per-agent in the Agent CR, not at the platform level. Create secrets in the agents namespace:

# Anthropic
kubectl create secret generic llm-api-keys \
  --namespace agents \
  --from-literal=ANTHROPIC_API_KEY=sk-ant-...

# OpenAI
kubectl create secret generic llm-api-keys \
  --namespace agents \
  --from-literal=OPENAI_API_KEY=sk-...

# Multiple providers in one secret
kubectl create secret generic llm-api-keys \
  --namespace agents \
  --from-literal=ANTHROPIC_API_KEY=sk-ant-... \
  --from-literal=OPENAI_API_KEY=sk-...

Agents reference providers via Provider CRs. Create one per backend:

# providers.yaml
apiVersion: agents.agentops.io/v1alpha1
kind: Provider
metadata:
  name: anthropic
  namespace: agents
spec:
  type: anthropic
  apiKeySecret:
    name: llm-api-keys
    key: ANTHROPIC_API_KEY
---
apiVersion: agents.agentops.io/v1alpha1
kind: Provider
metadata:
  name: openai
  namespace: agents
spec:
  type: openai
  apiKeySecret:
    name: llm-api-keys
    key: OPENAI_API_KEY

kubectl apply -f providers.yaml

Then reference them from agents via spec.providerRefs:

spec:
  providerRefs:
    - name: anthropic
    - name: openai

Image pull secrets

For private registries, configure global pull secrets:

# values-override.yaml
global:
  imagePullSecrets:
    - name: ghcr-pull-secret

Create the secret beforehand:

kubectl create secret docker-registry ghcr-pull-secret \
  --namespace agent-system \
  --docker-server=ghcr.io \
  --docker-username=YOUR_USER \
  --docker-password=YOUR_PAT

Console ingress

Enable ingress for external access to the console:

# values-override.yaml
agentops-console:
  ingress:
    enabled: true
    className: nginx
    hosts:
      - host: agentops.example.com
        paths:
          - path: /
            pathType: Prefix
    tls:
      - secretName: agentops-tls
        hosts:
          - agentops.example.com

Console environment (Tempo + Memory URLs)

The console BFF needs to know where Tempo and the memory service live. These depend on your Helm release name and namespace. Set them explicitly:

# values-override.yaml
agentops-console:
  env:
    - name: TEMPO_URL
      value: "http://agentops-agentops-platform-tempo.agent-system.svc.cluster.local:3200"
    - name: ENGRAM_URL_OVERRIDE
      value: "http://agentops-agentops-platform-memory.agents.svc.cluster.local:7437"

Memory service

Configure persistence size and image tag:

# values-override.yaml
memory:
  image:
    tag: "0.2.0"
  persistence:
    size: 5Gi
    storageClassName: "local-path"

Agent namespace

Change the namespace where agent workloads are deployed:

# values-override.yaml
agentNamespace: my-agents
createNamespace: true

Helm values reference

Key values for the umbrella chart:

Value	Default	Description
`agentNamespace`	`agents`	Namespace for agent workloads
`createNamespace`	`true`	Create the agent namespace
`global.imagePullSecrets`	`[]`	Image pull secrets for all components
Operator
`agentops-operator.enabled`	`true`	Deploy the operator
`agentops-operator.image.repository`	`ghcr.io/samyn92/agentops-operator`	Operator image
`agentops-operator.image.tag`	`""` (chart appVersion)	Operator image tag
Console
`agentops-console.enabled`	`true`	Deploy the console
`agentops-console.image.repository`	`ghcr.io/samyn92/agentops-console`	Console image
`agentops-console.image.tag`	`""` (chart appVersion)	Console image tag
`agentops-console.ingress.enabled`	`false`	Enable console ingress
`agentops-console.ingress.className`	`""`	Ingress class name
`agentops-console.ingress.hosts`	`[{host: agentops.local, ...}]`	Ingress host configuration
`agentops-console.service.type`	`ClusterIP`	Console service type
`agentops-console.service.port`	`80`	Console service port
`agentops-console.console.namespace`	`""`	Restrict console to single namespace
Memory
`memory.enabled`	`true`	Deploy the memory service
`memory.image.repository`	`ghcr.io/samyn92/agentops-memory`	Memory image
`memory.image.tag`	`0.2.0`	Memory image tag
`memory.persistence.size`	`1Gi`	SQLite PVC size
`memory.persistence.storageClassName`	`""` (cluster default)	Storage class
Tempo
`tempo.enabled`	`true`	Deploy Grafana Tempo
`tempo.tempo.retention`	`72h`	Trace retention period
`tempo.persistence.size`	`5Gi`	Tempo storage PVC size

Verifying the installation

Check all pods

kubectl get pods -n agent-system
kubectl get pods -n agents

Verify CRDs are installed

kubectl get crds | grep agentops

Expected output:

agents.agents.agentops.io              2026-01-01T00:00:00Z
agentruns.agents.agentops.io           2026-01-01T00:00:00Z
agenttools.agents.agentops.io          2026-01-01T00:00:00Z
agentresources.agents.agentops.io      2026-01-01T00:00:00Z
channels.agents.agentops.io            2026-01-01T00:00:00Z

Verify the memory service

kubectl get pods -n agents -l app=agentops-memory
kubectl logs -n agents -l app=agentops-memory --tail=5

Test console connectivity

kubectl port-forward svc/agentops-agentops-console -n agent-system 8080:80
# Open http://localhost:8080

Upgrading

helm upgrade agentops oci://ghcr.io/samyn92/charts/agentops-platform \
  --namespace agent-system \
  -f values-override.yaml

CRDs are upgraded automatically by the operator subchart. Existing agents continue running — the operator reconciles them against the new version.

Uninstalling

# Remove all agents and tools first
kubectl delete agents,agentruns,agenttools,agentresources,channels --all -n agents

# Uninstall the platform
helm uninstall agentops --namespace agent-system

# Clean up namespaces (optional)
kubectl delete namespace agent-system
kubectl delete namespace agents

# Remove CRDs (optional — this deletes all AgentOps resources cluster-wide)
kubectl delete crds \
  agents.agents.agentops.io \
  agentruns.agents.agentops.io \
  agenttools.agents.agentops.io \
  agentresources.agents.agentops.io \
  channels.agents.agentops.io

Warning

Deleting CRDs removes all AgentOps custom resources across the entire cluster, including any PVCs created by daemon agents. Back up any data you need before proceeding.