Analysis
Reason Machines has open-sourced Aspen, a self-hosted Git server built for AI agent traffic, independently implementing the design OpenAI described for Pando. Where a classic server queues every push to a repository together and builds a fresh pack for every fetch, Aspen scales writes per branch. It speaks plain Git over HTTPS and SSH, with packs in S3-compatible object storage and branch history in PostgreSQL.
The motivating workload is new: an agent pushes every few minutes and CI fetches after each push, so server work grows with clients rather than with new code. MarkTechPost reports the company's own test sustaining 567 pushes per second on one 8-core server — a vendor figure, not an independent benchmark.
The same report is candid about immaturity: no releases, no web UI, and a 24-hour durability gate not yet met. Aspen is an architecture worth studying now and a production dependency to adopt later, in that order.
Why it matters
If agents write to your repositories, measure push and fetch load per branch today. Adopt per-branch scaling when queuing, not storage, becomes the bottleneck — and wait for Aspen's durability gate before trusting it with the only copy of anything.
