Git Hosting Is Breaking Under AI Load — Entire Wants to Fix That
Entire’s new distributed Git network for AI coding agents is a regional mirroring layer that lets autonomous tools clone, fetch, and push against in‑region copies of GitHub repositories, offloading heavy traffic from the central host and reshaping version control as agent‑first infrastructure instead of a service optimized around human developers. Thomas Dohmke, who stepped down as GitHub’s CEO to start Entire, has opened a waitlisted preview of this Git network, pitched as a way to keep fleets of AI agents from overwhelming a single server and as a future competitor to GitHub’s core hosting. The move is unapologetically opinionated: Git hosting should look more like high‑frequency distributed systems and less like a shared folder for occasional human commits. In that sense, Entire is not a mere GitHub alternative; it is a bet that the center of gravity in developer tooling has already shifted toward AI coding infrastructure.
Inside Entire’s Git Network for AI Agents
Entire’s preview is deliberately narrow but strategically placed: it mirrors existing GitHub repositories into regional cells, then lets agents talk to those mirrors via an entire:// remote instead of hammering GitHub directly. The code still lives on GitHub; Entire’s Git network AI agents layer is a high‑volume read buffer that serves clones, fetches, commits, files, diffs, and merge bases close to where agents run. For mirrored branches, pushes go back to the upstream repository, so GitHub’s branch protections and checks remain in control. Technically, Entire rebuilt the server side of Git from scratch and describes its system as a Git‑compatible, distributed version control network with a global control plane for identity and placement and regional data planes for Git storage. One quotable benchmark claim stands out: Entire says its network sustained about 570,000 clones per hour under test.
Why Agentic Workflows Demand New AI Coding Infrastructure
The timing is not accidental. Agentic coding has changed repository traffic patterns so much that traditional centralized Git hosts are creaking. A human developer might clone once, pull occasionally, and push when work is ready; an agent swarm clones, inspects, checkpoints, and pushes in tight loops across many concurrent sessions, turning Git into a high‑frequency access layer. GitHub has acknowledged this rising load and has wrestled with availability incidents as coding agents and so‑called vibe coding overrun its main servers. Entire’s narrative is blunt: centralized Git hosting has become a constraint in “the era of agents,” showing up as rate limits, high latency and outages. From that perspective, AI coding infrastructure should not be a bolt‑on to human‑centric version control; it needs a Git network AI agents can treat as close, reliable storage, with regional proximity and fewer bottlenecks.
From Human-Centric GitHub to Agent-First Alternatives
Entire’s ambitions go far beyond mirroring. The company raised a USD 60 million (approx. RM280 million) seed round in February when it had only 15 employees, a scale described as unusually large for developer tools, and is already valued at a reported USD 300 million (approx. RM1.4 billion). It started with agent‑session memory — storing prompts, tool calls, and checkpoints alongside repository history — so reviewers can see why code changed, not just what changed. The Git network now extends that AI coding infrastructure down into the transport layer. Roadmap promises include native public and private repositories that no longer depend on GitHub, self‑hosted nodes, an open‑source Git backend, tamper‑evident history, and more decentralized operation. That is a clear GitHub alternative posture, even if the preview today is more offload helper than replacement.
What This Shift Means for Everyday Developers
For ordinary developers, the most immediate prize is less friction. Agents and humans can push, pull, and clone in‑region, closer to where they work, while still staying part of a global collaborative Git network. That can reduce latency, rate‑limit headaches, and the feeling of being beholden to a single US‑centric host that may go down mid‑day. At the same time, the preview is constrained: mirrored branches still write through to GitHub, and core features like native repositories, self‑hosted nodes, CI/CD and full decentralization are future roadmap items, not present reality. Developers should treat Entire as an opinionated layer for agent traffic, not yet as a wholesale migration path. The deeper message, though, is clear. The common theme in newer platforms is that Git hosting is evolving from casual human collaboration into infrastructure designed for AI agents operating at machine scale. Those who ignore that shift risk building workflows on top of infrastructure that is already out of step with how code is written.
