LANGUAGES Signal 443
Cursor's Origin moves Git source of truth to S3 with local NVMe as warm cache
↗
Cursor built a Git repository service called Origin that uses S3 object storage as the canonical source of truth via a write-ahead log, while local NVMe disks handle latency-sensitive operations like DAG traversal.
GitHub's approach of keeping three synchronized NVMe copies of every repository scales poorly as replica synchronization slows and Git resists eventual consistency. Cursor's architecture sidesteps both problems by making S3 the canonical store and treating local repositories as disposable warm caches, which matters increasingly as AI agents generate massive numbers of repositories.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Cursor's Origin service uses S3 as the source of truth through a write-ahead log, with local NVMe repositories acting as warm caches for latency-sensitive Git operations.
GitHub's Spokes architecture synchronizes three NVMe copies per repository, but synchronization latency grows with replica count and Git handles eventual consistency poorly.
AI agents exacerbate Git scaling problems by creating vast numbers of small, often throwaway repositories outside monorepos.
THE CLUSTER