Generative AI Google Brain

search-gemini-enterprise-gemini-enterprise-session

3 min read

Gemini Enterprise custom agent with Vertex AI session

Source notebook

Repo path: search/gemini-enterprise/gemini_enterprise_session.ipynb · Open on GitHub · advanced

Builds a Gemini Enterprise travel agent using ADK sub-agents and persistent Vertex AI sessions.

Summary

This notebook demonstrates a custom travel-planning agent integrated with Gemini Enterprise. It creates ADK sub-agents for query completeness checking and itinerary generation, runs them locally with VertexAiSessionService, updates an Agent Engine instance with an AdkApp, tests remote session-aware queries, and registers the agent through Discovery Engine API calls.

Key code patterns

Initialize Vertex AI

vertexai.init(
    project=PROJECT_ID,
    location=LOCATION,
    staging_bucket=STAGING_BUCKET,
)

Sets project, region, and staging bucket before creating Agent Engine resources.

Create ADK sub-agents

travel_query_validator = Agent(
    model="gemini-2.5-flash",
    name="travel_query_validator",
    instruction=QUERY_CHECK_PROMPT,
)
planner_agent = Agent(
    model="gemini-2.5-flash",
    name="planner_agent",
    instruction=ITINERARY_PROMPT,
)

Separates query validation and itinerary generation into focused agents.

Wrap sub-agents as tools

root_agent = Agent(
    name="root",
    model="gemini-2.5-flash",
    generate_content_config=types.GenerateContentConfig(temperature=0.01),
    tools=[AgentTool(agent=travel_query_validator), AgentTool(agent=planner_agent)],
)

Lets the root agent call specialist ADK agents through AgentTool.

Use persistent Vertex AI sessions

session_service = VertexAiSessionService(
    project=PROJECT_ID, location=LOCATION, agent_engine_id=app_name
)
runner = adk.Runner(
    agent=root_agent, app_name=app_name, session_service=session_service
)

Connects local ADK execution to Vertex AI session state so follow-up queries can use prior context.

Deploy AdkApp to Agent Engine

adk_app = reasoning_engines.AdkApp(
    agent=root_agent,
    enable_tracing=True,
    session_service_builder=session_service_builder,
)
remote_app = agent_engines.update(
    resource_name=agent_engine.name,
    agent_engine=adk_app,
)

Packages the ADK agent with a session service builder and updates an Agent Engine resource.

Models & APIs used

  • Models: gemini-2.5-flash
  • APIs / services: Vertex AI, Agent Engine, Discovery Engine
  • SDKs / libraries: google-adk, google-genai, google-cloud-aiplatform, vertexai

When to use this

Use this pattern when a Gemini Enterprise custom agent needs persistent conversation context across multi-turn workflows.

Gotchas & caveats

  • Project ID, project number, location, staging bucket, user ID, display name, app ID, client ID, client secret, and reasoning engine ID must be filled in.
  • The notebook writes a local .env file and sets GOOGLE_GENAI_USE_VERTEXAI=1 for Vertex AI backend use.
  • VertexAiSessionService requires an Agent Engine ID derived from a created Agent Engine resource.
  • The remote deployment pins google-adk to version 1.5.0 in requirements.
  • Gemini Enterprise registration uses gcloud access tokens and Discovery Engine v1alpha endpoints.
  • The registration example display name and description mention SQL generation even though the notebook builds a travel agent.

Best practices

  • Use VertexAiSessionService when an agent needs persistent session state.
  • Split agent responsibilities into query completeness checking and itinerary generation.
  • Use a low temperature of 0.01 for the root agent configuration.
  • Use a session_service_builder when creating reasoning_engines.AdkApp.
  • Test the same multi-turn queries locally and through the remote Agent Engine app.

Graph View

Backlinks