Quick Start¶

Deploy a Claude Code, DeepAgents, OpenClaw, or OpenCode agent in under 5 minutes.

Prerequisites¶

A cluster with Language Operator installed and access to a large language model.

Step 1: Create a Cluster¶

A LanguageCluster is a managed namespace for logically grouped agents, models, and tools.

kubectl apply -f - <<EOF
apiVersion: langop.io/v1alpha1
kind: LanguageCluster
metadata:
  name: language-operator-demo
spec:
  domain: agents.example.com
EOF

Wait for it to be ready:

kubectl wait languagecluster/language-operator-demo \
  --for=condition=Ready --timeout=60s

Switch into its namespace:

kubectl config set-context --current --namespace=language-operator-demo

Step 2: Configure an LLM¶

Store your API key in a secret:

kubectl create secret generic anthropic-credentials \
  --from-literal=api-key=sk-ant-your-key-here

Create a LanguageModel pointing to it:

kubectl apply -f - <<EOF
apiVersion: langop.io/v1alpha1
kind: LanguageModel
metadata:
  name: claude-sonnet
spec:
  provider: anthropic
  modelName: claude-sonnet-4-5
  apiKeySecretRef:
    name: anthropic-credentials
    key: api-key
EOF

Step 3: Deploy an Agent¶

Choose one of the bundled runtimes:

Claude CodeDeepAgentsOpenClawOpenCode

Claude Code authenticates to Anthropic directly (not through the gateway), so it doesn't use the LanguageModel from Step 2 — and it needs egress to reach api.anthropic.com:

kubectl apply -f - <<EOF
apiVersion: langop.io/v1alpha1
kind: LanguageAgent
metadata:
  name: code-agent
spec:
  runtime: claude-code
  networkPolicies:
    egress:
      - to:
          - cidr: "0.0.0.0/0"
        ports:
          - port: 443
            protocol: TCP
EOF

You'll log in interactively in Step 5 (or set an API key — see the Claude Code guide).

DeepAgents runs a task autonomously, so it needs instructions alongside the model:

kubectl apply -f - <<EOF
apiVersion: langop.io/v1alpha1
kind: LanguageAgent
metadata:
  name: researcher
spec:
  runtime: deepagents
  models:
    - name: claude-sonnet
  instructions: |
    Research the public API of the "deepagents" Python library, then write
    a concise summary to /workspace/summary.md. When the file is written, stop.
EOF

kubectl apply -f - <<EOF
apiVersion: langop.io/v1alpha1
kind: LanguageAgent
metadata:
  name: openclaw
spec:
  runtime: openclaw
  models:
    - name: claude-sonnet
EOF

kubectl apply -f - <<EOF
apiVersion: langop.io/v1alpha1
kind: LanguageAgent
metadata:
  name: opencode
spec:
  runtime: opencode
  models:
    - name: claude-sonnet
EOF

Step 4: Check Status¶

kubectl get languageagents

Watch the agent pod come up:

kubectl get pods -w

Step 5: Access the Agent¶

Claude CodeDeepAgentsOpenClawOpenCode

Port-forward the WebSocket terminal:

kubectl port-forward svc/code-agent 8080:8080

Open http://localhost:8080, then run /login inside Claude Code and complete the browser flow to authenticate with your Claude account. Credentials persist on the workspace PVC. See the Claude Code guide for headless API-key / token auth.

DeepAgents starts working as soon as the pod is Running — there's nothing to log into. Watch the run stream:

kubectl logs -f deploy/researcher

For a browser view of the same stream plus human-in-the-loop controls, port-forward the service:

kubectl port-forward svc/researcher 8080:8080
# open http://localhost:8080

Retrieve the auto-generated gateway token:

TOKEN=$(kubectl get secret openclaw-runtime -o jsonpath='{.data.OPENCLAW_GATEWAY_TOKEN}' | base64 -d)
echo "Token: $TOKEN"

If you have a domain configured on your LanguageCluster, open https://openclaw.<cluster-domain> and enter the token when prompted, or connect the OpenClaw CLI directly to wss://openclaw.<cluster-domain>.

Otherwise, port-forward for local access:

kubectl port-forward svc/openclaw 18789:18789
# connect to ws://localhost:18789 with the token above

OpenClaw uses a WebSocket gateway on port 18789. Connect using the OpenClaw browser extension or CLI client (see github.com/openclaw/openclaw).

Access is gated by the cluster OIDC proxy — there is no separate opencode password.

If you have a domain and auth enabled on your LanguageCluster, open https://opencode.<cluster-domain> and sign in through the cluster's OIDC provider.

Otherwise, port-forward for local access (this bypasses the proxy, so no login is required):

kubectl port-forward svc/opencode 3000:3000

Then open http://localhost:3000 or attach the TUI (OpenCode v1.0.10+):

opencode attach http://localhost:3000

Note: opencode attach against the https://opencode.<cluster-domain> URL won't work — the OIDC proxy needs a browser session. Attach via the port-forwarded http://localhost:3000 instead. opencode has no built-in auth, so enable cluster auth to protect it.