OpenAI's Pando Cuts Git Push Latency 5x for AI-Scale Development
As Codex enabled engineers to work in parallel, OpenAI’s monorepo infrastructure became a bottleneck. Push and fetch operations slowed, merge failures increased, and system reliability dropped below 90%. OpenAI built Pando, a new Git backend that redesigns server infrastructure while preserving protocol and client compatibility, allowing engineers and agents to work with unchanged tools and workflows.
Pando achieves scalability by independently coordinating branch updates—allowing parallel pushes to unrelated branches instead of serializing all writes—and reusing compressed object files across multiple fetches. These design choices deliver substantial gains: median push latency dropped nearly 5x, and the system now serves nearly a petabyte of Git data daily.
By maintaining Git protocol compatibility, OpenAI preserved existing workflows while enabling improvements developed for Pando to benefit the broader Git community.
Read the full article
Continue reading at Hacker News →This is an AI-generated summary. Read the original for the full story.