Devotd Github
NewAI agents

yuchenzhangtg/graft — a virtualized Git alternative for agents

yuchenzhangtg/graft is a Rust-based source control system designed to provide virtual workspaces and lazy loading for agents and monorepos.

Repositoryyuchenzhangtg/graft ↗

yuchenzhangtg/graft is a Rust-based source control system designed to provide virtual workspaces and lazy loading for agents and monorepos.

GitHub description: High-performance Git alternative for AI agents with instant (<30ms) virtual workspaces and AST merge. Inspired by Google CitC and Anthropic Scion.

stars
★ 7
Forks
0
Language
Rust
License
Apache-2.0
Read original source ↗View on GitHub ↗

yuchenzhangtg/graft is an alternative Git repository for AI agents and monorepos. There is no public latest release, and no separate version information is given. The repository description puts forward virtual workspace creation in under 30 milliseconds and AST merging. The documentation introduces a virtual file system that fetches the requested files and lazy loading.

Graft is a source control system that creates workspaces in a server-centric way and manages file contents by address. It is designed so that multiple agents handling the same repository do not have to download the entire history every time. AST merging is a way of analyzing code into a syntax tree and combining changes. The documentation explains a delta-centric storage flow, workspace creation, and a distributed storage structure, and also covers Rust builds and FUSE functionality. The storage configuration is divided into a local cache, optional cloud storage, and a metadata layer. According to the repository description, this project took inspiration from Google CitC and Anthropic Scion. The license is Apache-2.0.

The repository has 7 stars, and the momentum score in the research record is 2.8/10. The sub-30-millisecond workspace creation time is only a figure stated in the repository description; no comparative measurements or evaluation conditions are given. Therefore, it is hard to judge from the available material alone how effective concurrent work by multiple agents or AST merging is in real environments.