Google Adk Deploy

Google ADK — Deployment

Deployment Targets

Deploy to Cloud Run

Using ADK CLI

# Deploy directly (builds and deploys)
adk deploy cloud_run \
  --project my-gcp-project \
  --region us-central1 \
  --service_name my-agent-service \
  my_agent/

Manual Dockerfile

FROM python:3.11-slim

WORKDIR /app

COPY pyproject.toml .
COPY my_agent/ ./my_agent/

RUN pip install google-adk

EXPOSE 8080

CMD ["adk", "api_server", "--port", "8080", "my_agent/"]

Build and Deploy

# Build container
gcloud builds submit --tag gcr.io/my-project/my-agent

# Deploy to Cloud Run
gcloud run deploy my-agent-service \
  --image gcr.io/my-project/my-agent \
  --region us-central1 \
  --allow-unauthenticated \
  --set-env-vars "GOOGLE_API_KEY=your-key"

Deploy to Vertex AI Agent Engine

Using ADK CLI

adk deploy agent_engine \
  --project my-gcp-project \
  --region us-central1 \
  --display-name "My Agent" \
  my_agent/

Programmatic Deployment

For programmatic Vertex AI deployment, use the Google Cloud AI Platform SDK:

from google.cloud import aiplatform

aiplatform.init(project="my-gcp-project", location="us-central1")

# Use the ADK CLI for the actual deployment
# Programmatic deployment is handled via adk deploy CLI

Deploy to GKE

Dockerfile

FROM python:3.11-slim

WORKDIR /app
COPY . .
RUN pip install google-adk

EXPOSE 8080
CMD ["adk", "api_server", "--host", "0.0.0.0", "--port", "8080", "."]

Kubernetes Deployment

apiVersion: apps/v1
kind: Deployment
metadata:
  name: adk-agent
spec:
  replicas: 3
  selector:
    matchLabels:
      app: adk-agent
  template:
    metadata:
      labels:
        app: adk-agent
    spec:
      containers:
        - name: agent
          image: gcr.io/my-project/my-agent:latest
          ports:
            - containerPort: 8080
          env:
            - name: GOOGLE_API_KEY
              valueFrom:
                secretKeyRef:
                  name: agent-secrets
                  key: api-key
          resources:
            requests:
              memory: "512Mi"
              cpu: "250m"
---
apiVersion: v1
kind: Service
metadata:
  name: adk-agent-service
spec:
  selector:
    app: adk-agent
  ports:
    - port: 80
      targetPort: 8080
  type: LoadBalancer

API Server (Production Mode)

The adk api_server command starts a FastAPI server:

# Local testing
adk api_server my_agent/ --port 8080

# Production (in container)
adk api_server --host 0.0.0.0 --port 8080 my_agent/

API Endpoints

Environment Variables for Deployment

# Gemini API (simple)
GOOGLE_API_KEY=your-api-key

# Vertex AI (service account)
GOOGLE_CLOUD_PROJECT=my-project
GOOGLE_CLOUD_LOCATION=us-central1
GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json

Production Checklist

• Use DatabaseSessionService or VertexAiSessionService (not InMemory)
• Set up proper authentication (API keys, IAM)
• Configure memory service for cross-session recall
• Set resource limits (CPU, memory)
• Enable health checks
• Set up logging/monitoring (OpenTelemetry)
• Use secrets manager for API keys (not env vars in plain text)
• Configure auto-scaling rules

Key Rules

• adk api_server is the production entry point (FastAPI-based)
• Cloud Run is simplest for serverless deployment
• Vertex AI Agent Engine is fully managed but less customizable
• Always use persistent session services in production
• Never expose GOOGLE_API_KEY in container images — use secrets

Related Skills

• google-adk-app — App pattern (production configuration)
• google-adk-scaffold — Project structure
• google-adk-a2a — Deploying A2A-compatible agents