Oak

About Oak

Oak is a version control system built specifically for repositories that coding agents work in, on the argument that Git's data model was designed for humans reading history rather than for fleets of agents snapshotting and reverting constantly. It installs with a single shell command and does not require changing your model, harness, or IDE — you point agents at Oak and the workflow is otherwise the same. The pitch is latency, and the site publishes a benchmark harness you can run yourself rather than only claiming numbers: on repositories with many small files an initial snapshot drops from about 29.7 seconds to 1.4 seconds, task snapshots on multi-gigabyte binaries from 5.2 seconds to 321 milliseconds, and agent setup (clone plus checkout) on a wide config refactor from 540 milliseconds to about 59. The team is unusually candid about where it loses: repo init on 50,000 files, process cold start, and uncached dirty status are all currently slower than Git, and those rows are published on the same page as the wins. Large files are content-addressed and deduped across versions, and lazy mounts over FSKit or FUSE mean an agent does not have to clone full history to open one path. There is an `oak export` path back to plain Git, so adopting it is not a one-way door.

Key Features

Tags