AI's Distributed Git Future
Former GitHub CEO's Entire Unveils Distributed Git Network to Boost AI Coding Agents
Key Points
In testing, Entire staged 200 simulated clients performing shallow-clones from Frankfurt (40%), Paris, London, and Dublin over about 3 minutes, and 128 simulated agents pushing 1–10 files (2 KB each) per push over 2 minutes, with each push targeting its own branch, demonstrating the network's high-concurrency read/write capability.
Entire plans to open-source its Git backend and the accompanying benchmark suite to accompany the published performance figures, enabling external scrutiny and community contributions.
The preview is active in regions including the US, EU, and Australia, with a longer-term goal to build a globally connected but regionally resident network that supports native hosting of public and private repositories.
"By design, Git was always meant to be distributed," a framing from Thomas Dohmke as he pitches a move away from centralized hosting toward a distributed, agent-friendly Git network.
Former GitHub CEO Thomas Dohmke has launched a preview of Entire, a distributed Git network built specifically for AI coding agents. The platform lets developers mirror their existing GitHub repos across regional nodes, with early benchmarks showing roughly 570,000 clones per hour and 586 pushes per second on a single repository, according to The Next Web.
The launch comes as AI coding agents are putting enormous strain on centralized Git infrastructure. "By design, Git was always meant to be distributed," Dohmke said, framing Entire as a necessary fix for an era when billions of agents may soon be reading and writing code simultaneously, devops.com reported.
Benchmarks Show Hundreds of Thousands of Clones Per Hour
Entire ran two key tests to prove out its network. In the first, 200 simulated clients performed shallow clones from servers in Frankfurt, Paris, London, and Dublin over about three minutes. Frankfurt handled 40% of the load. The result: roughly 570,000 clones per hour from one repository, according to SiliconAngle.
The second test pushed 128 simulated agents writing 1 to 10 files — each 2 KB — to their own branches over two minutes. That produced about 586 pushes per second. A mixed test combining both reads and writes hit around 470 operations per second, Let's Data Science reported.
Mirror First, Then Native Hosting and Open Source
Right now, Entire works as a mirror layer. Developers keep their code on GitHub and point Entire at those repos. Agents then clone from the nearest regional node instead of hammering a single central server. Active regions already include the US, EU, and Australia, The Next Web noted.
Longer term, Entire plans to support native hosting of both public and private repos. The company also intends to open-source its Git backend and the benchmark suite it used for testing. That move is meant to invite outside scrutiny and community input, according to devops.com.
Decentralization as a Fix for Data Residency Concerns
One of the bigger promises Entire is making is around data sovereignty. The company wants to build a globally connected network where code actually stays within a given region. For businesses in the EU or Australia, that means their code never has to leave local servers while still being part of one open network, SiliconAngle reported.
Dohmke argues that centralized Git hosting was never built for the agent era. A single popular repo could face thousands of simultaneous clones from automated agents running in parallel. Spreading that load across regional mirrors cuts rate-limit problems and speeds up agent workflows, according to Let's Data Science.
AI Coding Boom Forces a Rethink of Version Control Infrastructure
Entire is not building in a vacuum. The rise of AI coding agents — tools that autonomously write, test, and push code — is straining platforms built for human-paced workflows. Agents can clone a repo thousands of times per hour, a pattern no centralized host was designed to handle at scale, devops.com noted.
The shift mirrors what happened to compute and storage as cloud scaled up — centralized models eventually gave way to distributed ones. Entire is betting the same transition is coming to version control. Whether developers and enterprises follow Dohmke's lead will depend on how quickly the open-source backend and native hosting features arrive, according to The Next Web.
Sentiment
Mostly PositivePublishers
10
Articles
2
Reach
12
