Developer Rebuilds Anthropic's Scion in Go, Hitting 2,500 Lands per Minute

An independent open-source take on Anthropic's internal Scion gives teams a Git-compatible server built for swarms of coding agents landing changes in parallel.

·
·
·
Developer Rebuilds Anthropic's Scion in Go, Hitting 2,500 Lands per MinutePRO
  • gitai is an open-source Go implementation of Anthropic's internal Scion version control system, MIT-licensed.
  • Server-centric design: clients stay online, history and blame run server-side, changes land via compare-and-swap.
  • Benchmark shows 1,862-2,544 lands/min vs classic Git's 240-272, with p99 latency dropping from 6.3s to 1.5s.
  • Pluggable conflict resolver works with any LLM provider, deterministic tool, or none; server never calls a model.
  • Storage backed by SQLite/Postgres plus object storage (S3, GCS, Azure); server nodes are rebuildable caches.
  • Available now at feder-cr/git-ai, v0.2.0, 2.5k stars, not affiliated with Anthropic.

Anthropic recently published the architecture of Scion, its internal version control system for code produced by large fleets of coding agents, without releasing the implementation. Developer feder-cr responded with gitai, an MIT-licensed Go project that applies the same server-centric design. At publication, the repository had about 2,500 stars and 275 forks. It retains real Git repositories underneath, preserving compatibility with existing Git tooling.

Why agents jam classic Git

Anthropic says its average volume rose from roughly 2,500 merged code changes per day in May 2026 to more than 30,000 per day by September. Traditional Git workflows assume that each worker clones a repository, works locally for an extended period, and pushes occasionally. Coding agents can instead launch thousands of short-lived tasks that touch a few files and try to update the same branches within seconds. At that concurrency, full clones, repeated history walks, merge queues, and competing reference updates constrain throughput.

Scion centralizes those operations. According to Anthropic’s description, its client maintains a bounded object cache, writes working-copy files, and watches the filesystem. The service handles history queries, blame, commit creation, merges, and rebases. gitai follows that model, so delegated operations require a live connection and clients store less repository state locally.

Pro article

This story is for Pro members

You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.

Trending
  • No trending articles

Comments

avatar

Next Reads