OpenAI's Pando Git backend cuts median push latency by nearly 80%

OpenAI published an engineering post on November 2 describing Pando, a new Git-compatible backend that now powers all development in the company's monorepo. Engineers and agents keep using the same Git commands and clients; what changed is the server side that stores and serves the repository.

The trigger was Codex. OpenAI says the size and demands on its monorepo grew at a staggering pace over the past year, because Codex lets engineers do far more work in parallel. That means more branches and pushes, more merges, and more fetches and clones from development environments, research jobs and CI. From around the end of last year, the existing system struggled: slow fetches and pushes, failed merges, and reliability that at times fell below 90%.

Pando targets two dimensions at once. It coordinates updates to each branch independently so pushes to unrelated branches proceed in parallel, and it reuses packfiles across fetches instead of generating a packfile for every client. A reader may download some objects it already has, but the server can hand the same bytes to many clients with little computation. The tradeoff is that a reader looking at several branches can see them at different points in time.

The production numbers are the headline: median push latency fell by nearly 80%, roughly a 5x improvement, and the system now delivers close to a petabyte of Git data per day. It has served more than 57 terabytes of Git packs in a single hour, with more than 99% of those bytes coming from local disk cache, on just a handful of CPU cores.

The migration was deliberately uneventful. OpenAI kept Pando and the old system in sync so traffic could move over without downtime, shipped its public Git distribution to developer machines and inside the ChatGPT desktop app, and migrated hundreds of integrations that depended on legacy APIs, webhooks and repository downloads. Codex itself helped: the team wrote skills that let agents guide developers through moving their environments and services.

OpenAI also frames the work as a contribution back to Git. Building on the Git protocol surfaced edge cases in concurrent downloads, cleanup after interrupted transfers and resuming completed downloads, and those fixes went upstream. Its public Git build can download and index independent packs in parallel and adds concepts such as explicit haves; several improvements are already in upstream Git, with more in OpenAI's public repository.

The signal here is bigger than one company's infrastructure. Agent-scale software production pushes source control past assumptions Git was designed around, and Pando suggests protocol compatibility and horizontal scale do not have to be mutually exclusive. OpenAI says it is exploring other internal uses, including workloads where large numbers of agents collaborate on a shared repository, and ways to share the technology with users — the parts worth watching next.