Entire launches distributed Git network for AI agents

Entire has launched a preview of a distributed Git network for software repositories, designed to let developers mirror existing GitHub repositories across regions.

The preview is available through a waitlist and includes active regions in the US, EU and Australia. Developers can mirror a GitHub repository in one step, keep the original codebase on GitHub, and direct agents to clone and pull from an Entire mirror instead.

The move targets growing strain on centralised code hosting as software teams use more AI agents to read, copy and update repositories at scale. Entire says the model is meant to reduce the rate limits, latency and outages that can emerge when large volumes of automated traffic rely on a single provider.

Founded by former GitHub Chief Executive Officer Thomas Dohmke, Entire is positioning the network as an alternative layer for Git-based development rather than an immediate replacement for existing hosts. Native hosting for new public and private repositories has not yet launched, but is planned as the company moves towards a broader decentralised and open source model.

Performance claims

To support the launch, Entire released benchmark figures. It says it sustained about 570,000 Git clones an hour from a single repository, based on 200 simulated clients carrying out shallow clones from Frankfurt, Paris, London and Dublin over about three minutes.

For Git push operations, Entire says it sustained 586 pushes a second, or about 2.1 million an hour, to a single repository or branch. The test used 128 simulated agents pushing one to 10 files of 2 KB each over two minutes, with each agent writing to its own branch.

A mixed test combining clone and push operations reached about 470 operations a second on a single repository, according to Entire. The company says the test reflected a repeated agent workflow of a shallow clone followed by five pushes, with 128 simulated agents and median latency of about 50 to 60 milliseconds.

Entire also compared its results with Cursor Origin, citing a public claim of 81,000 pushes an hour. On that basis, it says its own push rate was up to 25 times higher than that rival claim.

Agent workflow

The launch comes as code repositories face a different pattern of demand from AI-assisted development. Instead of a handful of human developers cloning a project and making occasional updates, automated agents can generate repeated pull, clone and push requests across many parallel sessions.

Entire argues that this shift makes Git's distributed nature more relevant again. Its system mirrors repositories across multiple regions so developers and agents can operate closer to where they run while remaining connected to a wider network.

"By design, Git was always meant to be distributed. As Linus Torvalds put it in his 2007 Google Tech Talk: 'If you're not distributed, you're not worth using.' In the era of agents, centralized Git hosting has become a fundamental constraint, as the strain of billions of agents and developers hammering a central server shows up in the form of rate limits, high latency, or even outages," said Thomas Dohmke, Chief Executive Officer of Entire.

He described the launch as the start of a broader effort to decentralise Git hosting.

"Today, we begin to return Git to its original promise, with a distributed, and soon fully decentralized and open-sourced network of interconnected nodes around the world. By doing so, we enable any developer or agent to host their code in-region, pushing, pulling, and cloning close to where they operate, fast and without bottlenecks, while still part of a global, collaborative network," Dohmke said.

Broader platform

The Git network is the latest part of Entire's wider developer platform. Since launching earlier this year, the company has offered what it calls a semantic memory layer that stores agent sessions, prompts, tool calls and checkpoints in the repository alongside the code.

Entire says it integrates with major coding agents including Claude Code, Codex, Cursor, Factory AI and GitHub Copilot. It has also introduced tools called Entire Blame, Entire Review, and code and semantic search, intended to help developers inspect not only what changed in code but also the reasoning and session history behind those changes.

Dohmke said those records are becoming central to software development as AI use grows.

"Session logs are now the second most important artifact in software development, and they belong in the repository alongside the code," Dohmke said. "With this semantic memory layer tied to the repo, agents stop repeating mistakes, which means higher accuracy, more productivity, and lower token spend. Developers can understand and verify what was built and why, which makes review far faster. And it opens up the possibility to build a new developer lifecycle, one that lets us understand and reason over the massive volumes of code AI agents now generate."

Entire says it has grown to more than 40 staff across the US, Australia, Germany, Spain, India, New Zealand, the UK, the Czech Republic and the Netherlands.