Entire GitHub Alternative: Trails Replaces Pull Requests

Thomas Dohmke ran GitHub for nearly four years. On Monday, October 26, the startup he founded after leaving it, Entire, opened a service that does most of what GitHub does, minus the pull request. Something called a Trail takes its place, though it can also sit on top of pull requests for projects mirrored from GitHub.

His reason is in the launch post on Entire’s site. The pull request, he wrote, “has become an audit log that only the agent reads.” Coding agents and review agents now trade long comments on the same thread, and the human who has to click merge gets no clear answer to the only question that matters, which is whether it’s safe.

What replaces it

Entire calls the replacement a Trail. It can start from a bug report, a feature request, or straight from a coding agent, and it can cover several repositories at once. Each time code is pushed, review agents run on their own in cloud sandboxes and each returns a score and a rationale. Entire’s post compares them to judges at an ice skating event.

The scores sit at the top of the page. Entire says many of its own repositories use three checks: Security, Slop, and Human Review Required. If the Slop score is orange, you tell your coding agent to fix it and push again until it goes green. The page shows the last five runs plus the current one, because, the company argues, a single pass often misses a serious problem.

A team can add “gates” that block a merge until a person approves or the tests pass. Reviews run on Claude Code and Codex, per Entire. Entire’s post adds that a failing Buildkite build, one of its two CI partners at launch, blocks the merge too.

The New Stack quotes Dohmke on how he uses it: if his internal “stop cop” check is green and confidence is high, “then we’re merging it.” Read that as the company’s own description of its own workflow, not as evidence that it’s safe to do.

The 89x figure

Entire says its Git server pushes up to 89 times faster than GitHub and six times faster than Cursor’s Origin. The chart in its post lists pushes per second, and dividing Entire’s 418 by GitHub’s 4.7 gives the 89.

Service	Pushes per second (Entire’s test)	Entire’s multiple
Entire	418	1x
Cursor	65	about 6x
Pierre	15	about 28x
GitLab	10	about 42x
GitHub	4.7	about 89x

Those are Entire’s numbers, from a benchmark it published itself. GadgetsNow’s coverage points out the same limit: it measures concurrent Git operations under chosen conditions, so a normal repository won’t necessarily see anything like that. I haven’t seen anyone outside the company repeat the test.

What it costs

The platform is free, according to The New Stack and GadgetsNow. Dohmke told The New Stack you pay for the tokens the review agents use, at market rates for whichever model you pick, once a $100 starter credit runs out (“right now,” per The New Stack). No rate card was published that I could find. That matters, because the pitch is more review loops per change, and every loop is a bill.

Hosting is live in the US, the EU, Australia and India. You can keep GitHub as the source of truth and mirror it, or move over. Public repositories and self-hosting are on the roadmap and GadgetsNow says no release dates were given.

Who’s asking, and when

Entire raised a $60 million seed round at a $300 million valuation in February, per Entire. One of the investors is M12, Microsoft’s venture arm, and Microsoft owns GitHub. In my view that’s an odd detail. The launch landed two days before GitHub’s own Universe conference, per GadgetsNow. GitHub already has agentic code review inside Copilot, as The New Stack notes. I’ll be watching what it announces on October 28 and 29.

My read

The problem is real. If agents write most of the code, nobody is reading diffs line by line, and pretending otherwise is the weak spot of the current pull request flow. Agents that work unattended, like OpenAI’s Dots, need a better answer to “who checked this” than a person skimming a long thread.

What I don’t buy yet is the fix. A score from one model on another model’s work is still a model’s opinion. Merging when the dashboard is green moves the trust from the author to the graders, and Entire’s graders are configurable, which means a team can quietly set the bar low. That’s my opinion, and it could be wrong. The evidence that would settle it is boring: how often a green Trail ships a bug. I haven’t found anyone who has published that.

Note on sources

I read Entire’s launch post in full. GeekWire, ITPro, The New Stack, ETEnterpriseAI and GadgetsNow are cited from the text I could retrieve of each. The Verge I only saw in a short excerpt, so I haven’t used it for any claim. The 89x, 71% token savings and other performance numbers are Entire’s claims.

GitHub's former CEO says the pull request is 'an audit log that only the agent reads.' His fix is a dashboard of AI scores.