OpenScion is an open-source reimplementation of Anthropic's Scion: a server-centric source control system engineered for very large monorepos where most clients are short-lived AI agents and CI jobs instead of full repository clones on developer laptops. It rejects Git’s offline-every-client model and instead runs history, blame, commits, rebases, and merges on the server; uses changesets that list only changed files; content-addresses blobs, trees, and changesets with BLAKE3; moves refs via compare-and-swap to avoid global locks; exposes source-control operations as RPCs; and fetches only the trees and blobs a client names. Those design choices target the scale Anthropic faced (tens of thousands of daily merges, a main-branch history near 100 GB, millions of daily checkouts) and yield client work that tracks the requested slice rather than repository size.
The implementation includes a layered architecture (per-node daemon, regional cache, server with metadata and object stores), a single protocol crate shared by client, server and GUI, benchmark tooling, and a command-line client. Synthetic benchmarks included show OpenScion importing and cloning several times faster than Git and one-directory clones that cost only the requested slice. The project is organized as Rust crates, has strong CI and coverage rules, provides quick-start commands, and is released under GPL-3.0.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.