Entire has launched a Git hosting platform that uses AI review agents to evaluate code changes, offering an alternative to conventional pull request workflows.
The platform introduces a review system called Trails, alongside distributed Git hosting and tools that preserve context from previous AI coding sessions. Entire was founded by former GitHub CEO Thomas Dohmke.
AI coding agents are accounting for a larger share of software development activity. According to GitHub’s 2026 Octoverse report, more than 40% of pull requests in September began with an agent’s first draft, up from 17% in March. GitHub recorded 110 million agent-started pull requests in September alone.
GitHub also found that 87% of agent-started pull requests modified multiple files, compared with 79% of pull requests without detected agent involvement.
Research from developer analytics company Faros AI also identified longer code review cycles. Its study of engineering telemetry from 22,000 developers across 4,000 teams found that median review time was approximately five times as high when comparing periods of lower and higher AI adoption within organisations. The findings establish an association rather than a causal relationship.
Entire developed Trails to change how developers review code produced by AI agents. According to the company, traditional pull requests organise code changes and review comments into chronological discussions, which can become difficult to follow when coding and review agents repeatedly exchange feedback.
“The pull request has become an audit log that only the agent reads,” Dohmke said in a blog post.
Changing how developers review code
Trails replaces the conventional pull request interface with automated evaluations that help developers assess code changes before merging them. A Trail can begin with a bug report, feature request, or coding agent, and can include changes across multiple repositories.
When code is pushed, configured review agents independently examine the changes in isolated cloud environments. Each agent produces an evaluation score and supporting findings, which are displayed in the Trail interface.
Rather than relying primarily on chronological comment threads, Trails displays evaluation scores and unresolved findings at the top of the review interface. Developers can compare the latest evaluation with the five previous runs to track whether issues have been addressed across successive code revisions.
The platform includes predefined review agents for code quality, security, and unwanted coding patterns. Developers can enable these agents for individual repositories or configure additional evaluations through the user interface or JSON configuration files.
Entire uses Claude Code and OpenAI Codex for these evaluations, with agents running in isolated environments provided by Daytona and E2B.
Developers can also configure agents to examine specific engineering requirements. In a technical article, Entire demonstrated how an agent reviews database migrations for destructive operations, table-locking risks, deployment compatibility, and rollback safety.
The agent assigns a risk score between zero and 100, identifies affected files and code lines, and provides explanations for its findings. Evaluations run when configured triggers and conditions are met, allowing developers to revise their code and compare the results of subsequent reviews.
Repository administrators can configure merge requirements through Gates, including human approvals, passing CI checks, and the resolution of review findings. Human approval is configurable rather than mandatory across all repositories, allowing teams to determine which checks must be completed before merging code.
Review findings can also be marked as open, resolved, dismissed, or stale. Changes to code associated with an existing finding can cause it to become stale, requiring further review rather than automatically treating the issue as resolved.
Managing agent-generated code with Git and CI
Beyond code reviews, Entire has developed a Git hosting infrastructure that uses a distributed object store and a transactional reference database. The system remains compatible with the Git protocol while allowing repositories to be replicated across multiple regions.
According to Entire’s own benchmarks, its infrastructure handled up to 89 times as many Git pushes per second as GitHub and approximately six times as many as Cursor Origin. The figures are based on company testing and have not been independently verified.
Entire initially offered hosting in the United States, European Union, Australia, and India before expanding to the United Kingdom, Singapore, and Brazil in late October. Developers can select hosting regions for their repositories, including locations that meet their data residency requirements.
The platform also supports GitHub repository mirroring, allowing developers to retain GitHub as their primary repository while using Trails for code reviews. Under this arrangement, code pushes continue through GitHub, and developers can use Trails alongside existing pull requests without migrating their repositories.
Entire’s documentation states that approvals recorded through its platform do not replace GitHub approvals. GitHub’s branch protection requirements continue to apply, meaning changes must satisfy the relevant GitHub rules even after Entire’s review conditions have passed.
Entire also integrates with continuous integration (CI) providers, including Buildkite, Depot, Dagger, and RWX. Pipeline results are incorporated into repository-level merge requirements, allowing automated build and test checks to operate alongside code review evaluations.
Entire added support for Claude Fable 5.2 in custom Trails review agents. Developers can select the model when configuring agents to assess code quality, security issues, and other repository-specific requirements.
Preserving coding context across sessions
Entire has also introduced Continuous Memory to retain information from previous AI coding sessions. The system connects development activity with repository history, allowing coding agents to retrieve earlier decisions and implementation details when working on subsequent changes.
The system consists of two open-source tools, entire-graph and entire-brain, which operate locally and can be used together.
The entire-graph tool creates a searchable map of repository structures, including code definitions, dependencies, function callers, and related files. It uses tree-sitter to analyse code locally without requiring network requests or external model calls.
According to Entire’s internal testing, the tool reduced token consumption by up to 71% when coding agents retrieved information about a codebase. The company has not provided independently verified benchmark results for this claim.
The second component, entire-brain, indexes previous coding-agent sessions and connects them with the repository’s code structure. Agents can use this information to retrieve earlier implementation decisions, recorded constraints, and details about previous changes.
The Entire CLI also captures coding-agent sessions and checkpoints alongside Git history. Developers can connect entire-brain to Entire’s backend to share stored context across team members and agents.
Stored session information can be accessed through Trails, giving reviewers access to the prompts and development decisions associated with a code change.
Research examines alternatives to pull requests
Separate research has examined how software development teams are responding to AI coding agents. A paper titled What Will Replace Pull Requests? Rethinking Software Collaboration for AI Agents, reviewed 50 documents from 31 organisations and individuals discussing alternative collaboration workflows.
Researchers Christoph Treude of Singapore Management University and Sebastian Baltes of Heidelberg University examined approaches to code integration, review, accountability, and external contributions. Their study identified review capacity and the loss of development context as recurring problems discussed in the reviewed material.
The study found that most proposed or adopted approaches retained pull requests in some form, including adding automated review tools, attaching session records, or restricting external contributions.
Among the documents reviewed, the researchers identified only one reported practical implementation of a replacement workflow, from development tools company Zed. The research covered documents published through October 6, before Entire introduced its new Git forge.
Zed’s Delta allows developers to implement, review, and integrate changes through conversations with coding agents rather than conventional pull requests. According to the paper, its repository had recorded 570 changes from 33 contributors without using pull requests as of September 16, 2026.
The researchers found that none of the measurements in the reviewed documents established whether alternative collaboration mechanisms improved code review effectiveness or defect detection. They also found that none of the documents discussing lost development context tested whether preserving coding-session records helped reviewers identify defects.
The paper was published as a preprint and drew on practitioner and vendor documents rather than a representative sample of software projects. Its authors did not independently reproduce the performance measurements reported by vendors.
(Photo by Igor Omilaev)
See also: GitHub says AI agents now start over 40% of pull requests
Want to dive deeper into the tools and frameworks shaping modern development? Check out the AI & Big Data Expo, taking place in Amsterdam, California, and London. Explore cutting-edge sessions on machine learning, data pipelines, and next-gen AI applications. The event is part of TechEx and co-located with other leading technology events. Click here for more information.
Developer Tech News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
