Cursor leverages S3 object storage to overcome Git's scaling limits
Cursor's Origin service stores Git data in Amazon S3 and uses a local NVMe cache, aiming to solve Git's scalability challenges.
Cursor has created a Git-based hosting platform named Origin that relies on Amazon S3 object storage rather than the multi-replica NVMe approach used by GitHub. Principal systems engineer Vicent Martí explains that each push is written to a write-ahead log as an immutable object in S3, while a local NVMe disk holds a reference copy serving as a fast cache. By synchronizing only this local reference instead of a quorum of replicas, the system can accept pushes at the speed of the underlying disk.
This design avoids the extensive DAG traversal required by traditional Git servers, allowing most operations to be performed locally. Martí notes that repositories can reside anywhere, with the S3 log acting as the ultimate source of truth. Origin is now available in beta for users with paid Cursor plans, and its reliability will be judged by the absence of outages.
Why it matters
It offers a scalable Git hosting model that could improve performance and reliability for large development teams.
In this story
