Entire builds git network for agents

1/ Entire’s pitch is simple: if teams are going to use fleets of AI agents in production, they need something closer to Git for agent work — not just for source code, but for the artifacts, decisions and review trail agents leave behind. The company opened a preview of a distributed Git network on July 8, with active regions in the U.S., EU and Australia. (thenewstack.io) 2/ The immediate product is infrastructure. Entire says developers can mirror an existing GitHub repository onto its network in one step, so code remains on GitHub while agents clone and pull from a regional Entire mirror. The company says that is meant to reduce the strain that large numbers of coding agents put on a single centralized host. (entire.io) 3/ That matters because agent traffic looks different from human traffic. Thomas Dohmke, Entire’s founder and former GitHub CEO, told The New Stack that advanced developers already use many agents in parallel, while Entire said centralized Git hosting can become a bottleneck through rate limits, latency or outages when agents repeatedly clone, fetch and push. (thenewstack.io) 4/ The broader idea goes beyond faster clones. Entire has been arguing since its February launch that agent-era development needs provenance: a record of how work was produced, what prompts and reasoning were involved, what files changed, and how to rewind if an agent goes off course. Its open-source CLI captures agent sessions alongside commits inside a Git workflow. (thenewstack.io) 5/ That is where the “Git network for agents” framing gets more interesting. Traditional Git is good at showing code diffs and commit history. It is much weaker at showing the full chain behind agent-generated work: which agent acted, what it was asked to do, whether a human approved it, and what should be rolled back if the output is wrong. Entire is trying to make that metadata first-class. (thenewstack.io) 6/ In practice, think of this as version control plus workflow evidence. The company’s materials and outside coverage describe a system aimed at session logs, checkpoints, attribution and recovery — the operational layer around AI-generated changes, not just the code delta itself. That matches a gap many teams are running into as they move from demos to production agent workflows. (thenewstack.io) 7/ Entire is also making a scale argument. In its July 8 launch note, the company said it sustained about 570,000 clones an hour from a single repository, 586 pushes per second to a single repository or branch, and about 470 mixed clone-and-push operations per second in its tests. Those figures come from Entire’s own benchmarks, but they show what it wants buyers to focus on: agent concurrency. (entire.io) 8/ The company is not fully replacing GitHub yet. In preview, Entire is positioned as a mirror and traffic offload layer for existing GitHub repos. But Entire said it plans in coming months to support native hosting for new public and private repositories and to further decentralize the network. That would move it closer to competing with the repository platforms it currently mirrors. (entire.io) 9/ Dohmke’s background is part of why the launch is getting attention. He left GitHub in 2025, launched Entire in February 2026, and raised a $60 million seed round led by Felicis, with participation from Madrona, Basis Set and Microsoft’s M12, according to The New Stack. (thenewstack.io) 10/ The deeper takeaway: as AI agents produce more code, documents and other business artifacts, teams need the same controls they expect for software changes — history, diffs, approvals, attribution and rollback. Entire is betting that the missing infrastructure layer for agent work is not another model, but a distributed system for managing provenance and collaboration around what agents create. (thenewstack.io)