Supermemory AI Releases Filesystem for Agents (smfs)

What changed

Supermemory AI has introduced smfs, a filesystem engineered specifically for AI agents, offering advanced retrieval and automatic memory profiling. This tool exposes a Supermemory container as a local filesystem, enabling agents to interact with their memory using standard file operations such as reading, writing, and grep-ing. smfs supports a wide array of file types, including PDFs, images, and videos, and integrates state-of-the-art retrieval mechanisms.

The smfs tool offers two primary access flows. The first allows it to be mounted as a local directory, compatible with editors, scripts, and any tool that reads files on macOS, Linux, devcontainers, Codespaces, and Docker. The second flow provides a virtual bash tool, packaged as a TypeScript library (@supermemory/bash), designed for runtimes that lack a local filesystem, such as Cloudflare Workers, serverless functions, edge runtimes, and browser-based agents.

Installation is facilitated via a simple curl command: curl -fsSL https://smfs.ai/install | bash, which supports macOS and Linux. Users require a Supermemory API key to operate. The quickstart guide demonstrates mounting a Supermemory container, for example, smfs mount agent_memory, which creates a local directory where files can be edited, cat-ed, or grep-ed. Writes are uploaded to Supermemory in the background, and remote changes are pulled in every 30 seconds. Unmounting is done with smfs unmount agent_memory.

Files within the mount are categorized into "memory paths" which are processed by Supermemory's memory pipeline for semantic search, and other files which serve as plain durable storage. The --memory-paths flag allows users to define which directories or files should generate memories, or to disable memory generation entirely.

The smfs command-line interface includes functionalities for login, status checks, listing mounts, syncing, and importantly, semantic search via smfs grep "query" [path]. A key feature is the integration of semantic search directly into the standard grep command. After running smfs init once, grep commands executed within an smfs mount will automatically leverage Supermemory's semantic index for flagless queries, while flagged grep commands fall back to literal string matching.

For environments without a local filesystem, the @supermemory/bash package provides a run_bash tool. This tool enables agents to execute Unix commands and use an sgrep command for semantic search across the entire container. The project is built using Rust and supports Docker and devcontainer environments, with specific FUSE requirements for Linux containers.

Why it matters for builders

This release offers AI builders a powerful new abstraction for managing agent data. By presenting memory as a familiar filesystem, smfs lowers the barrier to integrating complex memory retrieval into agent workflows. The seamless integration of semantic search into the standard grep command is particularly noteworthy, allowing agents to leverage advanced search capabilities without needing to learn new commands or tools.

Furthermore, the dual access flows—local mount and virtual bash tool—ensure broad compatibility across diverse agent architectures and deployment environments. This flexibility empowers builders to implement persistent, semantically searchable memory solutions whether their agents run on a local machine, in a container, or in serverless edge environments.

Practical impact

AI builders can now integrate smfs into their agent projects to provide persistent, semantically searchable memory. For agents operating in environments with a local filesystem, mounting a Supermemory container allows direct interaction with memory through standard file operations and the enhanced grep command. This could be used to store and retrieve agent logs, user interactions, or knowledge bases.

For agents deployed in serverless or edge environments, the @supermemory/bash package offers a way to access the same memory capabilities. Builders can plug the run_bash tool into their agent's toolset, enabling commands like sgrep to query the agent's memory contextually. This is particularly useful for agents that need to recall past conversations or access specific information without a persistent local state.

Developers can explore the installation process via curl -fsSL https://smfs.ai/install | bash and experiment with mounting containers using smfs mount <container_tag>. For those working with cloud-native agents, integrating the @supermemory/bash package into their agent's toolset is a direct path to leveraging these capabilities.

Caveats and source limits

The provided source material details the functionality and installation of smfs but does not include specific benchmark results for its retrieval performance or comparisons against other memory solutions. The exact pricing for the Supermemory API, which is required to use smfs, is not specified in the provided excerpts. While the tool supports various file types, the effectiveness of semantic search on different media like images and videos is not elaborated upon. The latest release version mentioned is v0.0.5, indicating that the project is still in its early stages, and further development and stabilization can be expected.