- Former GitHub CEO Thomas Dohmke launched Entire with $60 million in seed funding, targeting the infrastructure gap between human-paced Git workflows and AI agent workloads running at machine speed.
- Entire’s distributed Git network claims 331 pushes per second for 16 concurrent agents versus 8 on GitHub in its own benchmarks, roughly 40 times faster, though independent testing has not confirmed those figures.
Thomas Dohmke spent years running GitHub, which means he knows exactly where centralized Git breaks under pressure. His new platform, Entire launched this week with $60 million in seed funding and a direct argument: the infrastructure that handles human developers at scale was never built for hundreds of concurrent AI agents hammering the same repository.
The Agent Code Bottleneck
Centralized Git hosting was optimized for episodic, human-paced code pushes. That assumption falls apart when AI coding agents work continuously, cloning and pushing at machine speed across dozens of parallel sessions. Rate limits get hit fast, latency climbs, and what looked like a productivity gain turns into an infrastructure problem.
Standard Git also has a context gap. A commit records what changed but not the prompt that drove the change, the reasoning steps the agent took, or which tools it called along the way. Teams debugging or auditing AI-generated code end up piecing that together from external logs and separate documents, a fragmented workflow that scales badly.
Entire’s Distributed Git for AI
Entire’s answer to the throughput problem is a distributed Git network that mirrors repositories across multiple geographic regions. Agents clone from a local mirror rather than hammering a single origin server, which distributes load and sidesteps rate limits. In the company’s own benchmarks, Entire sustained 331 pushes per second for 16 concurrent agents in a mixed workload simulation, compared to 8 pushes per second on GitHub in the same test, roughly 40 times faster, per Entire’s figures. The company reports the simulation maintained around 470 clone-and-push operations per second overall. Independent testing has not confirmed those numbers, and real-world results will vary by workload.
The context problem gets its own solution: a semantic memory layer that records agent sessions, prompts, reasoning steps and tool calls alongside standard repository history. Two features sit on top of it. Entire Blame extends the familiar git blame function to link each line of code back to the specific agent conversation or prompt that produced it. Entire Review runs multi-agent diff analysis on pull requests, designed to speed up review of AI-generated contributions. The company plans to open source its Git backend and benchmarking tools.
Where Centralized Git Struggles
Dozens of agents repeatedly cloning, pulling and pushing to a single repository create the kind of sustained, concurrent load that centralized platforms were not designed for. Throttling follows, and teams find themselves managing infrastructure constraints rather than shipping code.
The lack of native semantic context compounds this. Enterprises building serious agentic workflows on standard Git end up bolting on external logging, storing prompts in separate systems and manually reconstructing the narrative behind automated changes. That overhead is real, and it gets heavier as agent usage grows. Gartner has warned that organizations routinely miscalculate what scaling AI actually costs; fragmented tooling of this kind is one contributing factor.
Scalability and Context: The Practical Difference
Entire’s distributed model returns to Git’s original decentralized design principles and applies them to agent-scale workloads. Mirroring globally reduces origin server pressure and cuts latency for geographically distributed agent fleets, a concrete architectural advantage over a single-point-of-truth setup when the request volume is coming from machines rather than people.
The semantic memory layer is the capability hardest to replicate with off-the-shelf tools, according to the company. Embedding the “why” directly into version history, rather than in a separate logging system, makes audit trails native to the development workflow. For enterprises where AI-generated code needs to meet compliance requirements or pass internal review, the company says that queryable provenance inside the repo matters. GitHub Copilot and similar tools augment human developers well, but they don’t solve the context problem for fully autonomous agents running without a human in the loop.
For teams already wrestling with agentic workflow costs, the recent CrewAI and LlamaIndex updates cutting agentic research costs are worth reading alongside this.
What Enterprises Should Watch
Entire is in preview, and the benchmark numbers come from the company itself, treat them as directional until independent testing exists. Felicis has called the round the largest seed investment ever for a developer tools startup, betting that agent-native infrastructure is a distinct market rather than a feature GitHub will eventually absorb. Whether that bet pays off depends on how fast enterprise agent deployments scale and whether Entire’s target pain points become widespread enough to justify a platform switch.
Entire has also announced plans to open source its Git backend, which the company says is intended to drive adoption of the distributed Git layer independent of the commercial platform. Teams should put Entire’s preview on the list where rate limits are already causing problems or where audit trails for agent decisions are a compliance requirement. For more on AI agents and automation tools, visit our AI Agents section.
