Ex-GitHub CEO's Entire previews distributed Git for AI agents

The interesting thing about SiliconANGLE's report on Entire is who is behind it. Thomas Dohmke ran GitHub. He is now selling the argument that the centralized Git host he used to run is the wrong shape for a world where AI agents, not humans, are the ones cloning and pushing at scale. His pitch, in his own words, is that "Git was always meant to be distributed" and that in the era of agents, centralized hosting shows up as "rate limits, high latency, or even outages."

The preview Entire opened this week is a distributed Git network operating by waitlist across the U.S., European Union, and Australia. On the company's own benchmarks, a single repository sustained about 570,000 clones an hour with 200 simulated clients shallow-cloning across Frankfurt, Paris, London, and Dublin, alongside 586 pushes a second, or roughly 2.1 million an hour, on a push test, and about 470 operations a second on a mixed workload. Take those specifics as reported, not settled. They are the company's own figures, and Entire says it plans to open-source both the Git backend and the benchmark suite so outsiders can rerun them.

The business is young and well capitalised. Entire launched in February with a $60 million seed round on a $300 million valuation and now employs more than 40 people across nine countries. Product-wise, it mirrors an existing GitHub repository in a single step while the canonical code stays on GitHub, and it plugs into coding agents including Claude Code, Codex, Cursor, Factory AI, and GitHub Copilot. There are two agent-native features on top: Entire Blame surfaces the agent session and prompt behind a line of code, and Entire Review sends a branch to multiple agents in parallel for an intent-aware review.

The honest caveats are the ones the reporting itself flags. The benchmarks are self-reported until the code lands, the mirroring model introduces a synchronization surface between Entire and GitHub with its own failure modes, and what the coverage does not give you is pricing, SLAs, how writes reconcile with the upstream host, or where the prompt history that Entire Blame captures actually lives.

The forward-looking part is the strategic bet, not the throughput number. If AI coding agents show up at the volume the industry keeps forecasting, the central Git host becomes a choke point rather than a moat, and the vendors best positioned are the ones offering a regional substrate that agent fleets can hammer without waking a rate limiter. The person making that bet used to run the choke point.