SQLite-based memory service implementation #2976
Description
Title: Feature Request: Add Persistent Local Memory Session with SQLite
Is your feature request related to a problem? Please describe.
Currently, the ADK's memory session is ephemeral. All contextual memory is lost once an agent session concludes. Any agent requiring long-term memory or statefulness cannot be effectively built. The only other option being vertex ai based one, which is not feasible in my case.
Describe the solution you'd like 💡
I propose the implementation of a local, persistent memory provider using SQLite. This would allow an agent's memory to be saved to a local .db file and reloaded in subsequent sessions, enabling true persistence without requiring external database infrastructure.
How it would work:
A new SQLiteMemory class or service would be introduced.
Users could enable it via a simple configuration, specifying a path to the database file.
The service would automatically handle the serialization and deserialization of the agent's memory to and from the SQLite database.
This functionality would be distinct from the existing database_session_service, which appears to have a different purpose. The new component would focus solely on persisting the agent's conversational state and learned context.