Can I pass user_id from root agent to remote a2a agent?

[thanab-gscalejp]

I want to use memory bank on remote a2a agent deployed by agent engine.

it will be useful only when the remote agents recognize the end users and be able to query memories for the users identified by "user_id".

i understand that memories are managed in user scope.

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: Yes, you can and should pass the user_id from a root agent to a remote A2A agent to ensure user-specific context, including memory, is correctly handled.

---

Hello! That's an excellent question. The Agent Development Kit (ADK) is designed to maintain user context across different agents, which is crucial for features like the memory bank.

When your root agent interacts with a remote agent (e.g., one deployed on Agent Engine), you can pass the user_id when creating a session or sending a query. This ensures that the session on the remote agent is correctly associated with the end-user.

Here is an example of how you would interact with a remote agent and pass the user_id [1]:

# When creating a session for a specific user
remote_session = await remote_app.async_create_session(user_id="USER_ID_FROM_ROOT")

# When sending a query within that user's session
async for event in remote_app.async_stream_query(
    user_id="USER_ID_FROM_ROOT",
    session_id=remote_session["id"],
    message="your message here",
):
    print(event)

By passing the user_id, the remote agent's session will be scoped to that user. Consequently, when the remote agent uses the memory bank (e.g., VertexAiMemoryBankService), it will correctly query and store memories for that specific user, just as you intended [2].

The ADK's internal mechanics for Agent-to-Agent (A2A) communication also rely on a context ID that is composed of the app name, user ID, and session ID, ensuring that the user's context is propagated correctly between agents [3, 4].

[1] Deploy to Vertex AI Agent Engine: https://google.github.io/adk-docs/deploy/agent-engine/
[2] Memory: Long-Term Knowledge with MemoryService: https://google.github.io/adk-docs/sessions/memory/
[3] google_search_agent_tool.py example: https://github.com/google/adk-python/blob/main/src/google/adk/tools/google_search_agent_tool.py
[4] A2A Context Utils: https://github.com/google/adk-python/blob/main/src/google/adk/a2a/converters/utils.py

[thanab-gscalejp]

it seems to be source code to call remote root agent not a2a agent.

but no problem.

i will ask this question in "https://github.com/a2aproject/a2a-python" instead.