supermemoryai/smfs

What it solves

Supermemory container storage, which is typically accessed via API, is exposed as a local filesystem. This allows users and AI agents to interact with their memory data using standard Unix tools (like ls, cat, and grep) and editors without needing to write custom API integration code.

How it works

The project provides two primary access methods:

1. Filesystem Mount: For environments with a kernel (macOS, Linux, Docker), it mounts a Supermemory container as a real local directory using FUSE or NFS. Changes are synced bidirectionally: local writes upload to the cloud, and remote changes are pulled every 30 seconds.
2. Virtual Bash Tool: For serverless or edge environments (Cloudflare Workers, browser agents) that lack a local filesystem, it provides a TypeScript package (@supermemory/bash) that simulates a bash environment, allowing agents to execute Unix commands and semantic searches.

Additionally, it includes a shell wrapper that overrides the standard grep command when inside a mount to perform semantic searches via Supermemory's index instead of literal string matching.

Who it’s for

• Developers who want to manage their AI memory containers as if they were local folders.
• AI Agent creators who want to give their agents a familiar Unix-like interface for reading, writing, and searching their long-term memory.

Highlights

• Semantic Grep: Transforms the standard grep command into a semantic search tool when used within a mount.
• Bidirectional Sync: Automatic background synchronization between the local mount and the Supermemory cloud.
• Flexible Runtimes: Supports both native OS mounts and a virtualized bash environment for serverless/edge functions.
• Configurable Memory Paths: Allows users to specify which directories or files within the mount are processed by the memory pipeline for semantic indexing.