question: how to use vector databases as memory?

[eyang-2]

Hi team, our company have vector database infrastructure what is the best way to integrate as a implementation under BaseMemoryService class? would maintainers accept a pr of implementation of the BaseMemoryService on vectordbs?

[GWeale]

Yes, BaseMemoryService is the right extension point for vector DB-backed memory.

Today, the recommended path is:

1. Implement a custom BaseMemoryService with:
   • add_session_to_memory(session)
   • search_memory(app_name, user_id, query) returning SearchMemoryResponse
2. Scope by app_name + user_id
3. Register it with a custom memory URI scheme (services.py or services.yaml) and run ADK with that --memory_service_uri.