Former GitHub CEO Launches Distributed Git Network for Agents

Agent fleets change the operational shape of source control. As autonomous coding agents scale, they create highly parallel, read-first workloads that stress centralized hosts' rate limits, latency, and availability; mirrored, regional caches and stronger provenance layers are a predictable emergence in response.

Entire, the developer-tools startup founded by former GitHub CEO Thomas Dohmke, launched a preview of a distributed Git network, according to coverage by SDTimes, ZDNet, TheNextWeb and the company's blog. DevOps and SDTimes report Entire raised a $60 million seed at a $300 million valuation earlier this year. The preview runs in regions across the US, Europe and Australia and is available via a waitlist, per the company announcement and press reporting. Entire published internal benchmarks it says demonstrate high-throughput agent workloads: roughly 570,000 repository clones per hour from a single repo and about 586 pushes per second, figures reported in SDTimes, ZDNet and TheNextWeb and attributed to Entire's tests. ZDNet also reports the company intends to open-source its rebuilt Git backend.

Git's native model is distributed at the object level, but commercial hosting centralized access patterns around single endpoints. Industry-pattern observations show that when many clients perform repeated shallow clones and pulls, bottlenecks move from storage to network egress, request rate limiting, and regional latency. Entire's approach, regional mirrors that serve clones and shallow pulls while leaving the canonical origin intact, aligns with common caching and content-distribution strategies used elsewhere in infrastructure. For practitioners, that implies tradeoffs to evaluate: cache consistency and freshness semantics for git fetch/git pull, authentication and authorization propagation across mirrors, and merge/conflict surface area when many agents push concurrently.

What Entire published (reported facts): The company blog and announcement describe a feature set that includes one-step mirroring of existing GitHub repositories onto Entire mirrors while preserving GitHub as the primary source of record, per Entire's blog post and press coverage. Entire's blog also argues for storing agent session logs and prompts alongside code as a "semantic memory" layer; that concept and rationale are presented in the company's blog post dated July 6, 2026. Benchmark details released by Entire and recapped by SDTimes and ZDNet show mixed clone+push tests sustaining roughly 470 clone+push operations/second on a single repository with median latency around 50-60 ms in simulated European-client scenarios.

Observed patterns from similar distributed systems suggest three areas teams should watch. First, provenance and audit: storing agent session logs alongside commits increases evidence for code origin and review, but raises storage, privacy, and access-control questions. Second, CI/CD and merge policies must be rethought for high-velocity agent pushes; feature-branch workflows and automated gating will see higher load. Third, networking and cost: mirrors reduce latency and egress spikes but add replication bandwidth and cross-region consistency costs. These are generic tradeoffs derivable from distributed-cache designs and are not claims about Entire's internal roadmap.

Watch real-world adoption and third-party validation of the published benchmarks; the numbers come from Entire's simulated-client tests and press reporting, not independent audits. Observe how popular coding agents and integrations behave with mirrored repos, reporting names include GitHub Copilot, Codex, and Claude Code as supported by Entire in press pieces, and whether agents maintain correct provenance when session logs are attached. Also monitor plans to open-source the backend, reported by ZDNet, because community review would let practitioners inspect consistency models, replication strategy, and security measures.

Closing note (reported fact): Multiple outlets report Entire employs more than 40 people across nine countries and expects to grow to roughly 60 employees by year-end, per DevOps coverage of the company's announcement.

Sources for reported facts include Entire's company blog and announcement, SDTimes, ZDNet, TheNextWeb, DevOps, and other contemporaneous press coverage. LDS analysis is explicitly labeled above and framed as industry patterns rather than statements about Entire's internal intentions.