Entire: Ex-GitHub CEO's Git Forge for Agents, Explained

§ 01What Entire shipped#

Entire’s launch post names three primitives, plus CI partners:

• Trails. A replacement for the pull request. Review agents score every push on evals and return findings, and the page shows those scores first.
• Distributed Git hosting. A new Git-compatible server with a distributed object store. You pick a primary region and replicate to others: the US, EU, Australia and India today.
• Continuous Memory. Agent sessions and prompts are synced into the repository and made searchable for the next agent.
• CI through partners. Buildkite and Depot at launch, each able to convert existing GitHub Actions workflows (bk pipeline convert, depot ci migrate).

§ 02How a Trail works#

Dohmke’s diagnosis of the pull request is blunt: “The pull request has become an audit log that only the agent reads.” Coding agents and review agents trade long comments in a linear timeline, and the human at the bottom still cannot tell whether the change is safe to merge.

A Trail turns that loop into numbers at the top of the page:

1. Start. A Trail opens from a bug report, a feature request or straight from your agent, and can span several repositories.
2. Score. On each push, review agents run independently in cloud sandboxes (Daytona and E2B, using Claude Code and Codex) and each returns a score and a rationale. The page shows the current run and the last five, so you can see whether scores have settled.
3. Close and merge. Findings carry severities and suggested fixes; agents or humans address or dismiss them. Per-repo Gates can require green evals, human approvals and passing CI before the merge.

Each project ships with preset agents for code quality, security and slop, and teams can define their own in the UI or in a .json file in the repo.

§ 03How fast, and where#

Entire says its forge is the fastest in production. Its own open-source benchmark, ForgeMark, measures pushes per second:

That is where “up to 89x faster than GitHub and 6x faster than Cursor Origin” comes from. It is a vendor benchmark; push throughput matters most when many agents push at once, less for a small team.

Leaving GitHub is optional. As Dohmke put it in a reply: “You don’t need to leave, you can have both, we even offer GitHub mirroring.” A mirrored repository keeps GitHub as the source of truth and can still run Trails on its pull requests; if GitHub goes down, Entire keeps serving clones from the regional mirror and agents can push to branches prefixed entire/unmirrored/. Moving for good uses the entire-migrate plugin, or a Detach button for mirrors hosted in the US.

§ 04Continuous Memory#

Dohmke’s thread makes the claim plainly: “Session logs are now the second most important artifact in software development, and they belong in the repo.”

Two open-source, local-first plugins for the Entire CLI do the work:

• entire-graph parses the repository locally with tree-sitter into a map of definitions, callers, types and routes. Entire says it makes no network requests or model calls, and saved up to 71% on tokens in its testing.
• entire-brain indexes recorded agent sessions against that map, so the next agent can recall why the code looks the way it does, which fixes already shipped, and which tests broke last time. Connected to Entire’s backend, those memories are shared across a team and shown on each Trail.

This builds on Checkpoints, the session-capture CLI Entire launched with in February. The CLI already integrates with Claude Code, Codex, Gemini CLI and, since October 16, Grok Build.

§ 05What it costs, and what is missing#

• Price. The New Stack reports the platform is free and customers pay only for the tokens the review agents use: a $100 starter credit, then market rates for the model you pick. Entire publishes no plan page yet.
• Repositories. Native hosting covers private repositories at launch; public repositories and self-hosted Git nodes are on the roadmap, per The New Stack.
• Open source. The CLI, entire-graph and entire-brain are MIT-licensed. The launch post does not describe the hosted forge or Trails as open source.

§ 06Who is behind it#

Entire announced a $60 million seed round at a $300 million valuation on February 10, 2026, led by Felicis with Madrona, M12, Basis Set, 20VC, Cherry Ventures, Picus Capital and Global Founders Capital. Angels included Gergely Orosz and Theo Browne. Entire says it built Entire using Entire.

§ 07Our conflict, and who this is for#

Our conflict, declared: CellCog sells AI employees, including an AI software engineer that keeps its own memory of a codebase between tasks, so we have a view on where agent memory should live. Treat our read with that in mind.

Entire’s bet is that review, not writing code, is now the bottleneck, and that the fix needs a new forge rather than an AI reviewer bolted onto a pull request. It fits engineering teams that already run several coding agents in parallel, want merge decisions backed by repeatable evals, and need code kept in a specific region. A team pushing a few human-written pull requests a day will feel less of the difference.

§ 08What we are watching#

• Pricing beyond the credit. Whether Entire publishes plans, and what review-agent runs cost per push at market rates.
• Public repositories. When open-source projects can host on Entire, and whether Dohmke’s hinted help for maintainers arrives.
• GitHub’s answer at Universe. What GitHub ships for agent review two days after this launch.
• Outside benchmarks. Independent push and clone numbers against ForgeMark’s.

§ 09Sources#

• Entire, The Git Forge for Agent-Human Collaboration, Thomas Dohmke, October 26, 2026.
• Thomas Dohmke, launch thread on X, October 26, 2026, 17:00 UTC, and reply on mirroring.
• The New Stack, Ex-GitHub CEO launches a GitHub rival that reinvents the pull request, Frederic Lardinois, October 26, 2026.
• Entire, seed round announcement, February 10, 2026.
• Entire, Entire CLI on GitHub.