deep-agents-memory

Persistent memory for AI agents across sessions is achieved using StoreBackend, which saves state so it survives threads and sessions. This allows agents to recall previous interactions and maintain long-term context without data loss.

Routing agent memory between ephemeral and persistent storage is handled by a CompositeBackend, which applies routing rules to direct different data paths to StateBackend or StoreBackend. This hybrid approach optimizes both speed and durability.

Yes, you can perform filesystem operations within an agent's memory backend using FilesystemMiddleware. It supports common file operations including ls, read_file, write_file, edit_file, glob, and grep for path-based routing and management.

To prevent cross-thread data leakage with persistent agent memory, you must implement safeguards and proper routing rules for CompositeBackend. Establishing clear boundaries and security considerations ensures safe development and production configurations.

Ephemeral in-memory state via StateBackend is used for fast, scoped memory during active tasks, while persistent storage is needed for data that must survive across threads and sessions. Hybrid routing combines both for optimized performance.