About Smfs
SMFS, short for Supermemory File System, mounts a Supermemory container as a real directory so an AI agent can use its memory with ordinary shell commands. Instead of learning a vector database, managing keys and wiring an SDK, the agent runs ls, cat and grep against the mount, and writes to files become memories while reads become recall.
Inside a mount, grep is transparently upgraded to hybrid semantic search, so a query such as an authentication flow returns results by meaning, while grep with the fixed string flag falls back to standard matching. A virtual profile file returns a live digest of the user that is synthesized from every memory in the container each time it is read, not stored. Sync is bidirectional: the tool pulls on a cache miss, pushes changed writes, keeps a SQLite cache that survives restarts and allows offline reads, and retries with exponential backoff when the network is unreliable. The mount uses NFS on macOS and FUSE on Linux, needs no kernel extension, and ships as one binary written in Rust. It is dual licensed under MIT or Apache 2.0.
It is aimed at developers building agents that are already fluent in bash and who want persistent memory without extra infrastructure. It is built by Supermemory Labs.
The site publishes no pricing for SMFS. It works with a Supermemory container, and that service's plans are not listed on the SMFS page.