How to deploy Google ADK agents to production (Vertex AI Agent Engine, Cloud Run, GKE)?

Deploying ADK Agents to Production

Deployment Decision Matrix

1. Vertex AI Agent Engine

The recommended fully-managed option:

pythonCopy
from google.adk.agents import Agent

agent = Agent(
    name="prod_agent",
    model="gemini-2.5-flash",
    instruction="You are a customer support agent.",
    tools=[search_kb, create_ticket],
)

bashCopy
# Deploy via CLI
gcloud ai agents deploy \
    --project=my-project \
    --region=us-central1 \
    --display-name="Customer Support Agent" \
    --agent-dir=./my_agent

2. Cloud Run (Serverless)

pythonCopy
# Dockerfile
FROM python:3.12-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt
COPY . .
EXPOSE 8080
CMD ["adk", "api_server", "--port", "8080", "my_agent"]

bashCopy
# Build and push
docker build -t gcr.io/PROJECT/my-agent .
docker push gcr.io/PROJECT/my-agent

# Deploy
gcloud run deploy my-agent \
    --image gcr.io/PROJECT/my-agent \
    --platform managed \
    --region us-central1 \
    --memory 2Gi \
    --set-env-vars GOOGLE_API_KEY=your-key \
    --min-instances 1 \
    --max-instances 10

3. GKE (Kubernetes)

yamlCopy
# deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: adk-agent
spec:
  replicas: 3
  selector:
    matchLabels:
      app: adk-agent
  template:
    spec:
      containers:
      - name: agent
        image: gcr.io/PROJECT/my-agent:latest
        ports:
        - containerPort: 8080
        resources:
          requests:
            memory: "1Gi"
            cpu: "500m"
          limits:
            memory: "2Gi"
            cpu: "1000m"
        env:
        - name: GOOGLE_API_KEY
          valueFrom:
            secretKeyRef:
              name: api-keys
              key: google-api-key
---
apiVersion: v1
kind: Service
metadata:
  name: adk-agent-svc
spec:
  type: LoadBalancer
  ports:
  - port: 80
    targetPort: 8080
  selector:
    app: adk-agent

Production Checklist

• Use
  textCopy

  DatabaseSessionService

  or
  textCopy

  VertexAiSessionService

  (not in-memory)
• Set
  textCopy

  max_llm_calls

  in RunConfig to prevent runaway loops
• Enable HITL for destructive tool operations
• Add observability (Cloud Trace, AgentOps, or Phoenix)
• Store secrets in Secret Manager (not env vars)
• Set up evaluation CI/CD pipeline
• Configure auto-scaling and health checks
• Enable structured logging
• Add rate limiting for API endpoints

Learn more at Deployment.