OpenAI details Pando, an in-house Git backend built for Codex-scale parallel work that cut push latency nearly 5x
OpenAI published an engineering post on 2 November describing Pando, a Git-compatible backend it rebuilt for its internal monorepo. According to CellCog’s summary of the post, Codex lets engineers run many changes in parallel and the old system’s reliability at times fell below 90%; Pando cut median push latency nearly 5x, serves nearly a petabyte of Git data a day and was built largely by Codex agents in 10 weeks.

As summarised, commits landed on master in OpenAI’s monorepo grew 1,378% from the first quarter of 2024 to the third quarter of 2026, mostly during 2026. One engineer now has several changes ready for CI at once, and each push can set off several jobs fetching the same commit; from late 2025 the old system suffered slow fetches and pushes and failed merges, with reliability at times below 90%.
Pando keeps the Git protocol and client, so engineers and agents did not have to change their habits. It drops repository-wide consistent snapshots: each branch updates independently and pushes only conflict on the same branch. Every pushed packfile is stored in blob storage and reused for many clients instead of building a pack per client, with small packs merged into larger ones in the background. In OpenAI’s production figures, median push latency fell nearly 5x, it serves nearly a petabyte a day and more than 57 TB in a peak hour, with over 99% of bytes from cache.
OpenAI says Pando was built with Codex in 10 weeks from kickoff to production: one engineer designed the core, teams of agents implemented it and a handful of engineers deployed it, with production traffic since April 2026. It is a fresh Rust codebase backed by a Git conformance suite, deterministic simulation tests and Lean and TLA+ formal proofs written by agents using GPT-6 Astra.
Pando is internal for now. OpenAI says it is exploring ways to share the underlying technology with users but gives no product, date or price. What is public is OpenAI’s own Git client build, openai/git, and some fixes have gone upstream to Git. The openai.com post returned 403 to our direct fetch, so these details rely on CellCog’s summary.
Sources
- OpenAI’s Pando: How Codex Outgrew Git, Explained — CellCog
- openai/git — OpenAI(GitHub)