OpenAI has introduced Pando, a new backend for its Git repositories designed to handle the massive increase in concurrent development work driven by its SI tools, specifically Codex. The move highlights the growing infrastructure demands placed on source control systems by agentic SI workflows, where thousands of parallel branches and pushes can strain traditional Git architectures.
What Happened
Over the past year, the size and demands on OpenAI’s monorepo increased significantly as engineers and SI agents utilized Codex to perform more work in parallel. According to the company, the existing systems supporting the monorepo began to struggle, with reliability falling below 90% at times due to slow fetches, pushes, and failed merges. To address this without disrupting engineering workflows, OpenAI built Pando, a system that retains compatibility with the Git protocol and client while optimizing for horizontally scalable writes and maximum read throughput. In production, Pando cut median push latency by nearly 5x and now delivers nearly a petabyte of Git data per day. The architecture relies on a relaxed consistency model, known as causal consistency, to allow independent branch updates to proceed in parallel, avoiding the coordination bottlenecks found in three-phase commit or write-ahead log systems.
Why It Matters
The release of Pando underscores a critical shift in the SI industry: the need for infrastructure that can support agent-scale operations. Traditional source control systems were designed for human-paced development, but the rise of agentic SI, such as Codex, enables engineers to generate and test multiple code changes simultaneously. This creates a surge in concurrent pushes and fetches that can overwhelm standard Git servers. By optimizing for this parallelism, OpenAI is addressing the specific engineering challenges posed by high-volume SI agent activity. The company also noted that improvements to the client could be useful beyond OpenAI, with plans to contribute enhancements to the broader Git community.
The Bottom Line
Pando represents a significant infrastructure upgrade for OpenAI, enabling its engineering team and SI agents to maintain high velocity and reliability in a monorepo environment. By solving the scaling challenges of concurrent writes and reads, the system supports the intensive workflows required by modern SI tools. This development highlights the ongoing evolution of backend infrastructure to keep pace with the capabilities of current SI models and agents.