Skip to content

OpenAI builds Pando Git backend to scale monorepo

OpenAI

The system cut median push latency by nearly 80% and currently serves nearly a petabyte of Git data per day.

Key points

  • Pando supports parallel updates across different branches while maintaining sequential order for pushes to the same branch.
  • Fetches return links to pre-stored packfiles to deliver CDN-like read performance through caching across clients.
  • OpenAI contributed upstream fixes back to Git to handle cleanup after interrupted packfile downloads.

Why it matters

Engineering teams scaling high-frequency AI agent workflows can see how large labs redesign core version control infrastructure.

Sources