Sharing and Persisting State Across Sub-Agents in Google ADK (ToolContext/CallbackContext Usage)

Google Cloud Build with AI Generative AI & Foundational Models
,
1

Hi!

I’m building a system using the Google Agent Developer Kit (ADK) framework that involves multiple sub-agents and their associated tools. My goal is to enable shared memory across the entire session, such that:

  • Information retrieved or computed by one tool is accessible to other tools or agents later in the same session
  • Repeated work (e.g. term definitions) can be cached and reused
  • The overall system maintains consistent state across sub-agent boundaries

I’m using function-based tools with the ToolContext parameter and trying to write to tool_context.state. I’m encountering this error:

400 INVALID_ARGUMENT. {‘error’: {‘code’: 400, ‘message’: ‘Request contains an invalid argument.’, ‘status’: ‘INVALID_ARGUMENT’}}

Questions

  1. What are the recommended ways to share memory/state between tools and sub-agents within a session using Google ADK?
  2. How should I use ToolContext or CallbackContext to safely and persistently read/write session state?
  3. Are class-based tools preferred over function tools when interacting with shared session state?
  4. Could you provide any working code examples or architectural patterns that demonstrate shared memory across agents and tools (e.g., glossary terms, cached results, document metadata)?

Thanks!