OpenAI Reveals Pando, a New Git Backend for Its Monorepo That Cut Push Latency by About 80% and Serves Nearly a Petabyte a Day
We built Pando, a new Git backend for our monorepo, to handle more pushes and fetches without changing how our engineers and agents use Git. Pando cut push latency by ~80% and now serves nearly a petabyte of Git data a day. We can grow Git capacity alongside agent usage.
Context
Coding agents push and fetch far more code than human teams, and OpenAI says Pando lets its Git capacity grow with agent usage without changing how anyone works.
Git is the version control system most software teams use, and a monorepo keeps a whole company's code in one repository, which strains Git at large scale. OpenAI's own engineers and its coding agents both work in its monorepo. Greg Brockman said OpenAI may offer Pando to others and asked interested teams to email him.
Deep dive
OpenAI Developers (@OpenAIDevs) posted on the evening of Nov 2 ET that OpenAI built Pando, a new Git backend for its monorepo. According to the post, Pando handles more pushes and fetches without changing how OpenAI's engineers and agents use Git.
OpenAI says Pando cut push latency by about 80% and now serves nearly a petabyte of Git data a day. The company says it can now grow Git capacity alongside agent usage. The post passed 970,000 views.
A diagram in the attached video shows how it works. When a writer pushes a branch, Pando stores the packfile in blob storage and records the ref version in a store called Turtle. When a reader fetches, Pando selects the packs the reader needs and a service called Express serves them.
Greg Brockman (@gdb) quoted the announcement and said Pando is a Git backend that can scale to very large monorepos. He said OpenAI took the hard road of continuing to use Git, which let it switch Pando on with no change to the developer experience. He added that OpenAI is considering providing it to others and asked interested teams to contact him by email. Tibo (@thsottiaux) also quoted the post, saying the team was finally able to talk about the work and crediting @tnm and team.
OpenAI has not published a release date, pricing or code for outside use.
What to expect next
Watch for whether OpenAI turns Pando into an external product or open source project, as Brockman suggested it might. A longer technical write up with benchmarks would show how it compares with existing large scale Git hosting. As more coding agents commit code around the clock, other large engineering teams may face the same load problem and look for similar backends.