OpenAI details Pando, a Git-compatible backend for Codex-scale development

OpenAI says it built Pando, a Git-compatible backend designed for the scale of development that Codex makes possible, to support work in its monorepo without changing existing Git workflows. The company said the system now powers development inside OpenAI and is intended to handle higher concurrency across branches.

What OpenAI says Pando is designed to do

In its engineering article published on 2026-11-02, OpenAI said Pando has two principal goals: horizontally scalable writes across branches and high-throughput reads. The company said pushes to unrelated refs can proceed independently, while updates to an individual ref use an atomic compare-and-swap-like operation.

OpenAI also said the system uses background compaction to combine small packfiles into progressively larger covers, using an approach similar to log-structured merge-trees. The company said it has used covers up to L4. OpenAI said these design choices are intended to improve parallelism, merge reliability and code-publication speed.

Context and limitations

The article frames Pando as a response to increasing monorepo demands and the concurrent development enabled by Codex. The performance and reliability benefits described are company claims rather than independently benchmarked results in the article. The publication date in the OpenAI engineering newsroom is 2026-11-02.