AgentFS provides a specialized filesystem for AI agents, addressing the need for structured storage of agent state, files, and operational history. It targets AI developers and researchers by offering auditability, reproducibility, and portability, enabling easier debugging, state management, and deployment of AI agents through a unified SQLite database abstraction.

How It Works

AgentFS is built around a SQLite database that acts as the agent's entire runtime environment. It exposes three core interfaces: a POSIX-like filesystem for file operations, a key-value store for agent context, and a tool call audit trail. This approach, leveraging Turso's in-process SQL database, allows every agent action—file modifications, state changes, tool invocations—to be recorded and queried via SQL, facilitating deep auditability and debugging. The entire agent state is contained within a single database file, promoting straightforward portability and reproducibility through simple file copying for snapshots. An experimental Linux-compatible sandbox is also available for controlled agent execution.

Quick Start & Requirements

• CLI Installation: Initialize an agent filesystem with agentfs init.
• SDK Installation: Use npm install agentfs-sdk for programmatic access.
• Prerequisites: Node.js for the SDK. The experimental sandbox requires a Linux environment.
• Documentation: User Manual, Agent Filesystem Specification, SDK Examples are available.

Highlighted Details

• Auditability: All filesystem operations, state changes, and tool calls are logged in SQLite, enabling SQL-based querying of agent history for debugging and compliance.
• Reproducibility: Agent states can be snapshotted by copying the agent.db file, allowing for exact reproduction of execution environments.
• Portability: The entire agent runtime, including files and history, is contained within a single SQLite file, simplifying version control and deployment.
• Sandboxing: An experimental, Linux-compatible sandbox provides an isolated environment for running agents with controlled filesystem access.

Maintenance & Community

The provided README does not contain specific details regarding maintainers, community channels (like Discord or Slack), or a public roadmap.

Licensing & Compatibility

The project is released under the MIT License, which generally permits commercial use and integration into closed-source projects, subject to the license terms.

Limitations & Caveats

This software is currently in ALPHA status and is intended for development, testing, and experimentation only. It is not recommended for use with critical data until it reaches production readiness. The sandboxing feature is explicitly noted as experimental.