
Image: Entire
Why it mattersCode review is the slow part when agents write most of the code. If a reviewer can trust the eval scores, the merge no longer waits for a line-by-line read.
Thomas Dohmke, the former CEO of GitHub, launched his new company Entire as a full GitHub alternative. Entire's blog post introduces the service as "the first Git forge built specifically for agent-human collaboration". The New Stack reports that Dohmke's startup closed a $60 million seed round in February. The team spent a year building what came out today.
The pitch is that pull requests broke down when agents write most of the code. The agent and the review tools fill the comments with long back and forth. The human at the end has to read all of that to decide whether to merge. Entire replaces the pull request with something it calls a Trail. A Trail shows eval scores at the top of the page. Review agents calculate the scores on every push.
How a Trail works
Each push triggers review agents in isolated cloud sandboxes. Entire says it rents those sandboxes from Daytona and E2B. Each agent returns a score and a short reason. The default evals include Risk, Intent drift, Slop, Test coverage, Security and Human Review Required. The last five scores are kept for every eval. Entire says one run can miss a fault another run catches. A team needs several runs to trust the result.
A team can configure its own review agents in a JSON file in the repo. The agents run on Claude Code and OpenAI's Codex, with the latest Anthropic and OpenAI models. Entire adds Gates, which say what must pass before a merge. Gates can require eval thresholds, resolved agent findings, green CI checks, and human approvals.
The platform is free. Users pay only for the tokens the review agents use. The tokens cost market rates on the model the user picks. Entire gives every new user a $100 starter credit, Dohmke told The New Stack. He said the business model is the gap between the discount Entire can get and the market rate.
The Git server and the memory
Entire's second part is a new Git-compatible source control server. Entire says it pushes up to 89 times faster than GitHub and 6 times faster than Cursor's Origin. The number is from Entire's own open-source benchmark. Repositories can live in a primary region and be mirrored in others, which Entire says covers data residency rules. The regions on launch day are the United States, European Union, Australia and India. Public repositories and self-hosted Git nodes are still on the roadmap.
The third part is Continuous Memory. Entire syncs an AI agent's session into the Git repository. That includes prompts, decisions and tool calls, so a later session can read the history. The company says this becomes a codebase graph plus an agentic brain. The next agent queries that memory at the start of its task. GitHub Copilot is a competing offer with agent code review. Dohmke's pitch is different. The agent's intent and work become part of the review, next to the code. That changes what a review is for.
What this changes for a team
Entire supports native hosting for private repositories at launch. A team on GitHub can mirror a repository to Entire today. The team can run Trails on top of pull requests without migrating. The migration tool is a plug-in called entire-migrate. A mirrored repository hosted in the United States can be detached from GitHub through the repository settings.
The launch comes two days before GitHub Universe, GitHub's annual developer conference, which The New Stack notes in its reporting.
Source
- Primary source: Entire blog, "The Git Forge for Agent-Human Collaboration", by Thomas Dohmke, and Entire's home page.
- Reporting: The New Stack, "Ex-GitHub CEO launches a GitHub rival that reinvents the pull request", by Frederic Lardinois.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

