AgentFS is a filesystem designed for AI agents that provides storage abstractions tailored to agent state management and execution.
AgentFS addresses the challenge of managing agent state, history, and reproducibility by storing the entire agent runtime—files, state, and operation history—in a single SQLite database file. The tool records every file operation, tool call, and state change in the database, enabling developers to query an agent's complete execution history using SQL. This approach allows snapshots of agent state to be captured and restored at any point, supporting reproducibility, debugging, and what-if scenario testing.
Developers building AI agents who need auditability and the ability to reproduce exact execution states should consider AgentFS. The tool suits projects where understanding agent behavior through historical analysis is important, or where compliance requirements demand complete operation logging. The single-file SQLite storage model makes agent filesystems portable across machines and deployable to any system where Turso runs, and they can be checked into version control. The project provides SDKs for TypeScript, Python, and Rust, along with a CLI that supports mounting the filesystem via FUSE on Linux and NFS on macOS, as well as command-line access to files and action timelines.
The project maintains active CI workflows and publishes SDKs across multiple package ecosystems. Development activity shows consistent attention to the core codebase with regular updates to the specification and user documentation, including a comprehensive manual and examples for getting started with both CLI and SDK approaches.
