OpenAI launches its self-built Git backend, Pando, now managing the main code repository.

ChainCatcher report: Following Cursor and Anthropic, OpenAI has unveiled its proprietary Git backend, Pando, which now manages the company’s entire main code repository. Over the past year, Codex enabled engineers to deploy large numbers of agents simultaneously to write code, resulting in a surge in code branches, pushes, merges, and pull requests. OpenAI’s original system frequently stalled and experienced merge failures, with reliability dropping below 90%. A single engineer led the core architecture design, with multiple Codex agent teams contributing to implementation; Pando was deployed into production within 10 weeks. This backend allows concurrent processing of code pushes across different branches, eliminating the need for sequential queuing. When fetching code, it no longer generates individual data packets for each client but instead reuses existing Git packets and distributes them directly via caching. After the overhaul, median latency for code pushes decreased by nearly 80%, with nearly 1 PB of data distributed daily and a cache hit rate exceeding 99%. However, different branches no longer guarantee a consistent state at the same moment. Cursor’s Origin remains Git-based and primarily aims to replace GitHub as a code hosting service. Anthropic’s Scion has abandoned Git entirely, moving commits, merges, and history queries to the server side. OpenAI retains Git clients and protocols, redesigning only the backend—engineers and agents continue using their existing development workflows with minimal changes.