deep-agents-memory

You achieve persistent memory for deep agents by implementing a StoreBackend, which provides cross-session storage to retain data like user preferences even after threads end. This ensures long-term data persistence without losing state.

The best way to isolate ephemeral state is using a StateBackend, which provides temporary per-thread memory that automatically resets when the thread ends. This keeps task-specific data cleanly separated.

Yes, you can route specific paths to different storage backends using a CompositeBackend. This enables hybrid storage routing, directing ephemeral data to StateBackend and persistent data to StoreBackend simultaneously.

FilesystemMiddleware integrates file operations directly with configured memory backends to support memory-backed workflows. It allows agents to perform file operations while maintaining the safety boundaries and persistence rules of the underlying storage.

You need hybrid storage routing when maintaining user preferences across sessions while keeping temporary data isolated per task. CompositeBackend routes specific paths to appropriate ephemeral or persistent backends for this mixed requirement.