OpenAI Launches Pando Git Backend Cutting Push Latency by 80%
OpenAI has introduced Pando, a new Git backend enabling parallel pushes for engineers and Codex agents. By enforcing causal consistency per branch to eliminate coordination overhead, the system reduces median push latency by nearly 80% while serving close to 1 petabyte of data daily.
Related Entities
Full Report
▲ Marktechpost AI · 2026-11-03 05
▲ OpenAI just introduced Pando, a new Git backend that lets engineers and Codex agents push in parallel without touching how they use Git.
The key idea is to stop coordinating the whole repository. Pando orders updates per branch (causal consistency), so pushes to unrelated branches never wait on each other. It also stores every pushed packfile and reuses it for later fetches instead of building a fresh pack per client.
The result: median push latency down nearly 80%, and nearly a petabyte of Git data served per day from a handful of CPU cores. Express caches hit above 99%, and clones saturate 10 Gbps links on standard cloud VMs. It is a trade-off though: readers looking at several branches get no single point-in-time snapshot, and repeat branch pushes resend everything outside trunk (p99 pack under 3 MiB).
Built in Rust in 10 weeks with Codex agents, in production since April. The backend is not open source, but client fixes are going upstream to Git......
Full analysis:
