Ex-GitHub CEO Launches Entire Git Forge Built for Coding Agents

By Sergio De SimoneOriginal title: Entire Git Forge:面向编码智能体重构 Git 协作方式
78AI Score

Former GitHub CEO Thomas Dohmke has launched Entire, a distributed Git hosting platform designed for collaboration between human developers and AI coding agents, with push speeds up to 89x faster than GitHub. Its core is the trails mechanism, which replaces traditional PR timelines with multi-round review-agent scoring, plus continuous memory that lets agents reuse repository knowledge across sessions.

Full Report

Entire is a Git code hosting platform designed specifically for collaboration between human developers and AI coding agents. Its core argument is that traditional software development infrastructure, especially the PR model, is not suited to scenarios where AI agents rapidly generate, review, and iterate on code.

Entire's current CEO and former GitHub CEO Thomas Dohmke explains that GitHub and GitLab center on asynchronous code review and version control, which changed how people collaborate with one another. By contrast, multi-agent code hosting platforms center on asynchronous orchestration, version control, and auditing of AI agents working alongside humans.

Entire is built on a fully distributed architecture, with code push speeds up to 89 times faster than GitHub. Its core component is trails, which address the pain points that AI code agents introduce into the traditional PR development model:

The timeline of a merge request has become an audit log that only agents read, and it cannot give developers enough information to confidently judge whether it is safe to merge.

A trail surfaces the most important information at the top, presented as an evaluation score that is updated with every code push. Developers can use these scores to judge whether the PR has stabilized or whether the AI agent should continue coding and reviewing for another round of iteration:

Merging should happen only when confidence is high, risk is low, and all issues have been cleared.

A trail can be initiated manually from a defect ticket or a feature request, and can span multiple repositories. Whenever new code is pushed, a review agent running in a cloud sandbox evaluates the changes and returns a score and rationale based on code quality, security, slop (low-quality code), and whether human review is needed. For example, if the slop score is orange, the developer can ask the coding agent to improve the code and push new changes, triggering another round of iteration.

Each evaluation shows the results of the last 5 rounds along with the current result. We consider this critical, because a single evaluation often misses key issues. Iterating over multiple rounds until the score stabilizes is the key to successful agentic coding.

Evaluation agents can be configured either in the Entire platform UI or through repository-level JSON files.

Review agents can provide fix suggestions or optimization proposals, which remain visible while the trail is open. All issues found by reviews must be resolved or closed before merging. Developers can also configure repository-level Gates that define the conditions for a merge to pass.

Another foundational concept Entire introduces is continuous memory, which keeps all accumulated knowledge in the repository so that agents can reuse it across sessions instead of starting from scratch. Continuous memory is implemented by two open-source tools: one builds a precomputed graph of the Git repository, including ranked code search, definitions and callers, types, and routes; the other creates a persistent memory layer by combining the codebase graph with previous agent sessions and information recorded in Git history.

This ensures agents always get the right context at the right time — that is, the "why" behind the "what changed."

In addition, Entire supports integration with CI platforms such as Buildkite and Depot.

In the related Hacker News discussion, user conartist6 expressed the view that engineers have already delegated the work of writing code to agents, because programming languages and tools still leave much to be desired. Haimantika, who currently works at Entire, pointed out that the problem was never the language or the tools themselves:

The reasoning process is discarded between the person and the artifact. Commit messages have always been bad, documentation goes stale, and "why is this code here" can only be explained by whoever remembers. Agents have simply amplified this problem, because now no one remembers. This is exactly the problem we want to solve. If a better programming language in the future could also solve it, honestly, I would welcome that too.

Entire is just one of the new players in the source control space, alongside Zed DeltaDB, Anthropic Scion, East River Source Control, Cursor Origin, Radicle, and others.

[Statement: This article was translated by InfoQ and may not be reproduced without permission.]

View the original English article: https://www.infoq.com/news/2026/11/entire-git-forge-coding-agents/