- Preview Launch: Entire has opened a waitlisted preview of a regional GitHub mirroring network built for AI coding agents.
- What It Does: Agents can clone and fetch from regional Entire mirrors while GitHub remains the upstream repository for mirrored projects.
- Benchmark Caveat: Entire claims up to 570K clones per hour, but the results are company-run and the highest write-throughput path depends on Entire-native or unmirrored branches.
- Roadmap Boundary: Native repositories, self-hosted nodes, an open-source Git backend, and broader decentralization remain planned rather than generally available.
Entire, the developer-platform startup founded by former GitHub chief executive Thomas Dohmke, has opened a waitlisted preview for a regional Git hosting network designed for AI coding agents. The preview lets developers mirror existing GitHub repositories onto Entire so agents can clone and fetch from regional copies rather than repeatedly hitting the upstream host.
The launch matters because agentic coding changes the traffic pattern around repositories. A human developer may clone once, pull occasionally, and push when a branch is ready. An agent fleet can clone, inspect, checkpoint, and push repeatedly across many parallel sessions. GitHub has acknowledged that agentic development workflows have accelerated sharply since late 2025 and forced it to plan for far higher scale; separate reporting has tied that pressure to April reliability incidents and later cloud-capacity questions.
Dohmke has framed Entire’s distributed Git network for AI coding agents as a return to Git’s original design. “By design, Git was always meant to be decentralized,” he wrote in the launch post. The important technical question is narrower: how much of today’s agent-driven repository load can Entire move away from a central GitHub origin, and under what workloads?
What Entire’s Preview Actually Does
The preview starts with a mirror, not migration. A developer authorizes the Entire GitHub App, creates a mirror, and then gives humans or agents an entire:// remote for the selected region. Entire’s mirror workflow describes this as a synced, read-optimized copy of a GitHub repository; the team can keep its existing GitHub workflow while selected users and agents clone or fetch through Entire. The broader Entire documentation positions repository mirroring as the bridge between existing Git projects and Entire’s agent-context tools.
That distinction matters. Entire can absorb heavy read traffic from agents by serving clones, fetches, refs, commits, files, diffs, and merge bases from the chosen regional cell. But for ordinary mirrored branches, a push through the mirror is still forwarded to the upstream GitHub repository, and GitHub write access, branch protection rules, and required checks still apply. In other words, the preview’s clearest immediate benefit is read offload and regional proximity, not a full replacement for GitHub’s write path.
Under the hood, Entire describes the system as a Git-compatible repository network with a global control plane for identity and placement and regional data planes for Git object storage. The launch includes active preview regions in the United States, the European Union, and Australia.
The Git network also connects to Entire’s earlier agent-memory product. Entire stores agent sessions, prompts, tool calls, and checkpoints with repository history so reviewers can inspect why code changed, not only what changed. Its security and privacy documentation says redacted transcripts and checkpoint metadata are stored as Git data in a dedicated repository branch, while also warning that redaction is best-effort and that sensitive workloads should use private repositories.
Under the hood, Entire is a fully rebuilt Git backend, running on nodes distributed around the world. Users can pin their repos to a single region or spread the load across several. More regions coming, self-hosting too. See you soon India, Japan, Singapore, Brazil, and many more… pic.twitter.com/lRYeXmsl6x
— Thomas Dohmke (@ashtom) July 8, 2026
What the Benchmarks Say, and What They Do Not Prove
Entire’s published performance numbers give the launch its headline scale, but they should be read as vendor benchmarks rather than independent proof. The most useful way to read them is by workload type.
| Workload | Published result | Published test shape | What devs should watch |
|---|---|---|---|
| Clone throughput | 570K clones per hour | 200 simulated clients shallow-cloning a production repository across multiple European regions for about three minutes. | This supports the read-offload pitch, but it is still a short, company-run test with shallow clones and no independent reproduction yet. |
| Mixed clone-and-push workload | 1.7M operations per hour, or about 470 operations per second | 32 simulated agents repeatedly running a mixed workload of shallow clones and checkpoint pushes with roughly 50–60 ms median latency. | This is closer to an agent lifecycle, but real repositories, slower regions, larger histories, CI hooks, branch protections, and enterprise policies may change the result. |
| Push throughput | 2.1M pushes per hour, or 586 pushes per second | 128 simulated agents pushing small file changes to separate branches. | The highest write-throughput path is not the same as a normal GitHub-backed mirror push. Entire says branches prefixed with entire/unmirrored/ stay local to the region and power the direct benchmark path. |
Entire measured its write workloads with ForgeMark, an MIT-licensed benchmark tool for concurrent Git-push throughput and latency against smart-HTTP Git forges. That is better than a black-box marketing number because outside developers can inspect and run the tool. Still, the benchmark results remain Entire’s own until independent teams test them against their own repositories, regions, branch strategies, access rules, and agent workloads.
The central benchmark caveat is that is workload-specific. Entire’s launch post says ordinary GitHub-backed branches push only as fast as GitHub allows, while Entire-native or unmirrored branches can absorb much more concurrency. That means the preview may be immediately useful for heavy clone and fetch traffic, but the strongest write-throughput claims depend on branches that are not simply ordinary GitHub mirror writes.
Why Git Hosting Is Becoming an Agent-Infrastructure Problem
Entire’s timing is not accidental. GitHub’s own April availability update admitted that after the company introduced a 10X capacity plan in October 2025, by February 2026 needed to redesign for 30X scale because agentic development workflows were driving growth in repository creation, pull requests, API usage, automation, and large-repository workloads.
Entire is one response to that pressure, but not the only one. Cursor has positioned Origin as “a git forge for the agentic era.” Dohmke has also personally backed Tangled, a separate decentralized Git effort. The common theme is that AI agents make Git hosting less like occasional human collaboration and more like high-frequency distributed systems infrastructure.
Entire entered that race with unusually large early backing. Dohmke launched the company in February 2026 with a $60 million seed round and a reported $300 million valuation, initially focused on agent-session memory, review, and traceability. The Git mirror preview extends that earlier product into the repository transport layer.
What Is Still Missing
The preview leaves several enterprise questions open. Native public and private repositories on Entire, self-hosted nodes, an open-source Git backend, tamper-evident branch history, policy-as-code branch protection, CI/CD features, and broader decentralized operation are listed as roadmap items rather than generally available capabilities.
Teams evaluating the preview should therefore separate four questions: whether Entire improves clone and fetch speed for their agents; whether its write paths fit their branch and review model; whether its security and redaction model is acceptable for their codebase; and whether the roadmap arrives quickly enough to justify depending on the platform before native hosting and self-hosting are available.
That is also why ForgeMark matters now. The tool is already public, so developers can begin testing Git-push behavior on infrastructure they control. Entire’s planned open-source Git backend would be a separate milestone, because it would let outside teams inspect and operate more of the system behind the benchmark claims.
The Takeaway
Entire’s preview is a credible attempt to solve a real agent-era bottleneck: high-volume repository reads from many automated coding sessions. Its strongest near-term case is regional GitHub mirroring for clone and fetch traffic. Its biggest unproven claim is broader write scalability under real enterprise conditions.
The launch should therefore be read as a promising infrastructure preview, not a proven replacement for established Git hosts. Entire will have a stronger case once outside developers reproduce the benchmark results, test larger repositories across slower regions, and see whether the promised native hosting, self-hosting, and open-source backend arrive on the stated roadmap.
