Former GitHub CEO Unveils Distributed Git Network Built for AI Coding Agents
As AI coding agents become a larger part of software development, a new startup founded by former GitHub CEO Thomas Dohmke is introducing infrastructure built to address the growing pressure on centralized Git hosting.
The new company, Entire, has launched a preview of a distributed Git network that allows developers to mirror GitHub repositories across multiple geographic regions, enabling AI coding agents to clone and access code from nearby mirrors instead of repeatedly querying a single centralized repository.
The company says the approach is intended to reduce latency, avoid rate limits and improve reliability as software development faces the greater computing demand created by autonomous AI agents.
The preview is available through a waitlist and currently operates in the US, Europe and Australia. Developers can mirror an existing GitHub repository in a single step while continuing to use GitHub as the primary source of record. AI agents retrieve code from regional Entire mirrors, reducing the heavy read traffic generated by large numbers of concurrent coding agents.
Founded earlier this year, Entire started with a $60 million seed funding round that valued the company at $300 million. The fully remote company employs more than 40 people across nine countries and expects to grow to roughly 60 employees by the end of the year.
Long term, its roadmap includes native hosting for new public and private repositories and a fully decentralized network that allows companies to keep source code within specific geographic regions to meet sovereignty requirements while remaining connected through a global Git infrastructure.
Benchmarks Released
Dohmke touts Entire’s decentralized approach as a return to Git’s original architecture rather than a departure from it. Git was created as a distributed version control system, but commercial hosting platforms concentrated repositories on centralized infrastructure. The rapid expansion of AI agents has challenged the limits of that model, with developers encountering rate limits and service disruptions during heavy activity.
Entire released benchmark results intended to demonstrate the platform’s scalability. In internal testing, the company reported sustaining approximately 570,000 repository clones per hour from a single repository using simulated clients distributed across several European regions. Additional tests showed roughly 586 Git pushes per second to a single repository, equivalent to more than two million pushes per hour. A mixed workload simulating real-world AI agent behavior maintained approximately 470 clone-and-push operations per second.
The company said it plans to open source both its Git backend and the benchmarking tools used to generate those results, allowing independent validation.
The distributed network forms one component of Entire’s developer platform, which focuses on preserving the context behind AI-generated code. Rather than storing only source code, the platform records agent sessions, prompts, reasoning and tool calls alongside repository history.
Several new capabilities build on that semantic memory layer. For instance, Entire Blame extends the traditional Git blame function by identifying not only who modified a line of code but also the AI conversation and prompt that produced it. Entire Review enables multiple AI agents to review a code branch simultaneously using contextual information about development intent. Code and Semantic Search allows developers to search for why code was written, rather than simply tracing what changed over time.
