What problem does it solve?
Deep Agents often need both fast, thread-scoped working memory and durable, cross-session persistence, but teams struggle to set up consistent backend routing for files and memory operations.
Core Features & Use Cases
- Ephemeral thread memory (StateBackend): Keeps temporary working files only for the lifetime of a thread.
- Persistent cross-thread memory (StoreBackend): Maintains long-term storage across sessions and threads.
- Hybrid path routing (CompositeBackend): Routes different filesystem paths to different backends (ephemeral vs persistent) to match real workflows.
- FilesystemMiddleware tools: Enables common file operations like ls, read_file, write_file, edit_file, glob, and grep.
Use case: You want an agent to draft scratch files during a conversation, but permanently store user preferences under a dedicated folder so later threads can retrieve and reuse them.
Quick Start
Configure your Deep Agent to use CompositeBackend so paths under /memories/ persist while other paths remain ephemeral for the current thread.