Entire launches distributed Git network for AI coding agents

Entire, the developer-platform startup led by former GitHub CEO Thomas Dohmke, has unveiled a preview of a distributed Git network purpose-built for the age of AI coding agents. This new system is designed to spread code hosting across multiple regions rather than relying on a single central provider.

The logic behind this move is becoming increasingly familiar in the agent era. As more coding agents clone and pull code simultaneously, central hosts begin to strain under the load. GitHub itself felt this pressure when it had to freeze new Copilot sign-ups because agentic usage was breaking its economic model.

Entire’s solution is a mirroring approach. The preview, currently open via a waitlist with active regions in the United States, the European Union, and Australia, allows a developer to mirror an existing GitHub repository in a single step. The original code stays where it is, while agents clone and pull from a regional Entire copy instead.

The goal is to offload heavy, concurrent read traffic so agents can continue working without hitting rate limits. In the coming months, the company plans to let developers host new public and private repositories natively on the platform.

The longer-term vision involves fuller decentralization. Entire intends to build out a network of interconnected nodes that would allow teams to keep code in-region for data residency and sovereignty while remaining part of a single global system.

Dohmke frames this as a return to first principles. “By design, Git was always meant to be distributed,” he said, arguing that centralized hosting has become a fundamental constraint now that billions of agents and developers are hammering the same servers.

He is not alone in making that bet. Dohmke has personally backed Tangled, a separate decentralized Git effort, and the idea of routing around a single provider has been gathering momentum across the developer world.

To support its performance claims, Entire published a set of benchmarks from its own testing. The company says it sustained roughly 570,000 clones per hour from a single repository, 586 pushes per second, and about 470 combined clone-and-push operations per second on one repo, using scripts of simulated agents.

The mixed test is designed to mirror how an agent actually works: cloning a repository, pushing a handful of changes, and then repeating in a tight loop. Entire says it held that pattern at roughly 50 to 60 milliseconds of median latency.

These figures are the company’s own and have not been independently verified. Entire says it will open-source the Git backend and the benchmark suite behind them, allowing outsiders to test the numbers for themselves.

The network sits on top of the product Entire launched in February, a semantic memory layer meant to stop agents from repeating past mistakes. It now plugs into major coding tools including Claude Code, Codex, Cursor, Factory AI, and GitHub Copilot, storing each session, prompt, and tool call in the repository alongside the code.

New features build on that history. Entire Blame shows not just who changed a line but the agent session and prompt behind it. Entire Review sends a branch to several agents at once for an intent-aware read of the diff. A search feature allows developers and agents to query not just how code changed over time but why it was written.

Dohmke argues the logs matter as much as the code. “Session logs are now the second most important artifact in software development, and they belong in the repository alongside the code,” he said, casting the memory layer as a way to cut wasted tokens and speed up review.

The company has grown to more than 40 people spread across nine countries, with plans to reach 60 by the end of the year. It is doing so on the back of a record $60 million seed round, raised in February at a $300 million valuation, which gave the venture one of the largest launches in developer tools to date.