smfs

smfs (Supermemory File System) presents an approach to integrating memory and semantic search capabilities directly into the filesystem, allowing agents and developers to interact with AI memory as if it were a standard directory. By mounting a Supermemory container as a real directory, smfs enables users and agents to perform familiar file operations such as ls, cat, and grep, making memory management accessible through standard terminal commands. This design eliminates the need for agents to interact with complex vector databases, external SDKs, or specialized APIs, streamlining workflows by collapsing multiple memory-related services into a single mount point.

A notable feature is the semantic search upgrade to the grep command within the mount. When grep is used, it performs meaning-based searches across the contents, while grep -F retains standard exact-match behavior. The system also provides a live profile capability, where reading profile.md synthesizes a fresh digest from all stored memories each time it is accessed, rather than serving a static file. This live profile offers up-to-date context for agents, and writes to the filesystem are treated as new memories, with reads functioning as recall operations.

smfs supports real POSIX mounts, utilizing NFS on macOS and FUSE on Linux, and is implemented in Rust with strict safety measures that forbid unsafe code. It does not require kernel extensions or macFUSE. The tool is resilient to network issues, supporting offline reads and synchronization through a local SQLite cache and employing exponential backoff for syncing when connectivity is unstable. Files are cloud-synced within the Supermemory container, ensuring that all agents have access to the latest information.

The system is designed for a range of use cases, including legal document search, financial analysis, research and documentation, support and knowledge base management, HR operations, and personalized agent experiences. It allows agents to search across various document formats—such as PDFs, videos, screenshots, audio, and text—without requiring separate OCR, transcription, or parsing pipelines. smfs is available under MIT or Apache-2.0 licenses.