Sponsor

gitai

Open-source Git server built for many coding agents landing changes at once

gitai preview

About

Git designed for AI agents, based on the published Scion design

Classic Git hosting assumes a person clones, works for hours and pushes once. Fleets of coding agents instead land thousands of small changes on the same branch within seconds. gitai (the extensions are called Graft) starts from a fork of GitLab's Gitaly and adds sparse, always-online clients (`graft clone`, `graft sparse`), server-side blame, history and code search memoized per commit, queue-free landing with batched reference updates, pluggable automatic conflict resolution (any LLM provider, a deterministic tool or none), an ordered changeset log you can watch live, and a storage core on SQLite or PostgreSQL plus object storage so servers scale out. Version 0.1.0, first public release; interfaces may change before 1.0. Not affiliated with Anthropic. Building it needs Go, a C toolchain, Meson and Ninja. Repo: https://github.com/feder-cr/git-ai.

Features

  • Sparse clients that fetch only what they need
  • Server-side blame, history and search, cached per commit
  • Lands changes with compare-and-swap, no merge queue
  • Pluggable automatic conflict resolution, vendor neutral
  • Changeset log with the agent behind every change
  • SQLite or PostgreSQL plus S3, GCS or Azure storage
  • Compatible with existing Gitaly RPCs

Install

Related skills

All skills
All articles

More Coding

View all

Sourcery MCP

Let your coding agent request Sourcery reviews and fix the comments and security findings itself

Worktrunk

Git worktree CLI built for running several coding agents in parallel

DBHub

Token efficient database MCP server for Postgres, MySQL, SQL Server, Oracle and SQLite