#

Key takeaways

• OpenAI published an engineering account of Pando on November 2, 2026.

• Pando is a new backend serving Git repositories in OpenAI's monorepo.

• OpenAI says it adopted Pando without interrupting engineering work.

• The system targets horizontally scalable writes and high read throughput.

• OpenAI says it is exploring ways to share Pando technology.

What OpenAI says Pando is

OpenAI published an engineering account of Pando on November 2, 2026. The post was written by Aaron Friel, Taylor Blau, and Ted Nyman. Pando is a new backend serving Git repositories, and OpenAI says it now supports development in the company's monorepo.

The company says it adopted the system without interrupting engineering work. It also says engineers kept their familiar Git workflows. That makes compatibility a central part of the design, not an afterthought.

Why OpenAI built it

OpenAI says the design addresses greater parallel development associated with Codex. The post points to more branches and pushes, more merges, and more downloads from development environments, research jobs, and continuous integration. Those are the workload patterns the system is meant to handle.

OpenAI states two main goals for Pando. The first is horizontally scalable writes. The second is high read throughput. The post frames the system as a response to increasing repository activity inside OpenAI.

How the system works

Pando coordinates updates independently across branches. It also reuses packfiles across fetches. In practical terms, that means the server can deliver the same bytes to multiple clients with less computation.

The system trades repository-wide point-in-time consistency for causal consistency. OpenAI says that means a reader examining multiple branches may observe them at different moments. That is an important operational choice, because it favors scale and throughput over a single global snapshot.

OpenAI also says reused packs may include some objects a client already holds. That detail helps explain how the system reduces repeated work during fetches. The post presents this as part of the efficiency model behind Pando.

Compatibility with Git

OpenAI says Pando retains Git protocol and client compatibility. That is a significant design constraint, because it lets the system work with existing Git workflows. The company also describes improvements such as explicit haves.

According to the post, explicit haves let a server distinguish a branch tip from history a pushing client may omit. OpenAI says some improvements are upstream in Git, while others are in its public repository and patch tracker. The post presents this as a mix of shared and internal work.

Reported production results

OpenAI reports nearly fivefold lower median push latency and delivery of nearly a petabyte of Git data each day. The company says those figures describe its own production environment. They are not general benchmarks for every repository.

That distinction matters. The numbers show what Pando handled inside OpenAI, but they do not establish universal performance for other deployments. The post uses them as evidence of the system's scale in OpenAI's environment.

Sharing and availability

OpenAI says it is exploring ways to share Pando technology. The post does not establish general product availability or pricing. It only states that sharing is under exploration.

Editorial note

The source presents Pando as an internal engineering system with Git compatibility and scale-oriented design choices. The strongest supported claims are about OpenAI's own environment, its workflow continuity, and the system's technical approach. Any broader adoption or commercial availability would need separate confirmation from OpenAI.