Google ADK Advanced (Python)

Google ADK Advanced (Python)

Latest: 2.0.0b1 | Updated: April 2026 Upstream: https://github.com/google/adk-python | https://google.github.io/adk-docs/get-started/python/

Patterns

• Multi-agent coordination with runner
• HITL and escalation
• GCP deploy: Cloud Run, Vertex AI, tracing

Quickstart Pattern

Use LlmAgent (the parameter is singular instruction, not instructions) and wire everything through an InMemoryRunner for local development. Runner takes a single root agent= (or app=) — there is no agents= list parameter.

import asyncio
from google.adk.agents import LlmAgent
from google.adk.tools import AgentTool
from google.adk.runners import InMemoryRunner

researcher = LlmAgent(
    name="researcher",
    model="gemini-2.5-flash",
    instruction="Research the topic and cite sources.",
)
writer = LlmAgent(
    name="writer",
    model="gemini-2.5-flash",
    instruction="Draft and refine copy using the `researcher` tool.",
    tools=[AgentTool(agent=researcher)],
)

async def main():
    runner = InMemoryRunner(agent=writer, app_name="pipeline")
    events = await runner.run_debug(
        "Summarise LangGraph with references",
        user_id="u1",
        session_id="s1",
    )

asyncio.run(main())

See the agents page for the full constructor surface and the runner page for production wiring.

Deployment

• Package for Cloud Run; use Secret Manager for keys
• Enable Cloud Trace and Cloud Logging; export OTLP if needed

Cloud Run (service.yaml)

apiVersion: serving.knative.dev/v1
kind: Service
metadata:
  name: adk-agent
  namespace: default
spec:
  template:
    spec:
      containers:
        - image: gcr.io/PROJECT_ID/adk-agent:latest
          env:
            - name: GOOGLE_APPLICATION_CREDENTIALS
              value: /var/secrets/google/key.json

GitHub Actions (Cloud Run deploy)

name: deploy-cloud-run
on: { push: { branches: [ main ] } }
jobs:
  deploy:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: google-github-actions/auth@v2
        with:
          credentials_json: ${{ secrets.GCP_SA_KEY }}
      - uses: google-github-actions/setup-gcloud@v2
      - run: gcloud builds submit --tag gcr.io/$PROJECT_ID/adk-agent:latest .
      - run: gcloud run deploy adk-agent --image gcr.io/$PROJECT_ID/adk-agent:latest --region $REGION --platform managed

Security Best Practices

• Store keys in cloud secret managers (Key Vault/Secret Manager/Secrets Manager)
• Use least-privileged IAM roles; separate read/write roles for tools
• Network egress allowlisting for tool calls
• Encrypt logs; avoid sensitive data in traces
• Rotate keys regularly; monitor for anomalies