Runner override in adk web to use custom session and memory services

[webravolab]

Hi, my question is simple: Is there a way to use a custom session and memory service (i.e. VertexAiRagMemoryService) when using "adk web" as UI interface?
I think that I need some way to override the Runner object to inject my custom services, but I cannot find any way to do so.
Any help will be appreciated.

[dexhunter]

You can configure these services through adk web without overriding its Runner.

For VertexAiRagMemoryService, the current CLI provides a rag:// memory URI:

export GOOGLE_CLOUD_PROJECT="your-project"
export GOOGLE_CLOUD_LOCATION="your-rag-corpus-location"

adk web ./my_agents \
  --session_service_uri="sqlite:///./sessions.db" \
  --memory_service_uri="rag://YOUR_RAG_CORPUS_ID"

Use the corpus ID after rag://, not the full projects/.../ragCorpora/... resource name. The registered factory constructs that resource name from the project, location, and ID. Google Cloud credentials and corpus access must already be configured.

For your own session or memory class, put a services.py file directly inside my_agents/ and register factories with get_service_registry().register_session_service(...) or .register_memory_service(...). Then pass a URI using your registered scheme to the corresponding CLI option. ADK loads that file before constructing the services; the ServiceRegistry guide includes a complete example.

Configuring the memory service makes it available to the agent; memory ingestion and retrieval still need to be part of your application's flow.