Entire
Entire · Entire: Git Forge Where Review Agents Score Every Push, from Former GitHub CEO Thomas Dohmke
Entire is a Git forge built for teams whose code is mostly written by coding agents: Trails replace pull requests, review agents running Claude Code or Codex in cloud sandboxes score every push and file findings, and Gates decide when a change can merge. It suits AI-native teams that want agent review, agent session history, and fast regional Git hosting in one place, either natively or as a mirror of their GitHub repositories.
Best for
AI-native engineering teams that already ship most code through Claude Code, Codex, Cursor, or similar agents and want automated multi-agent review, explicit merge gates, and agent session history attached to every change, with fast regional Git hosting or a GitHub mirror
Not ideal for
Open-source maintainers who need public repositories, organizations that need self-hosted or on-premises Git or published compliance certifications, teams that want to keep GitHub Actions or another CI provider besides Buildkite and Depot without staying on a GitHub mirror, repositories that depend on Git LFS, and anyone looking for an agent that writes the code itself
Who it's for
Engineering teams and individual developers who run coding agents heavily and want a forge built around agent review, merge gates, and agent session history
Capabilities
- Trails replace pull requests: a Trail holds the goal and acceptance criteria, linked branches from one or more repositories, agent sessions, review findings, CI results, and human approvals, and can be created on entire.com or from the CLI
- Review agents run on every push to a Trail branch in isolated cloud sandboxes (Daytona and E2B), each returning a score (0-100 for the default monitors) with a rationale, and the Trail shows the current run beside the last five
- Preset review agents for code review, security review, and Slop (sloppy patterns and drift), enabled per repository
- Custom agents with your own instructions, a choice of harness (Claude Code or Codex), model, reasoning effort, timeout, score polarity, and triggers on push, manually, or on a schedule
- Findings with severity levels and suggested fixes that people or coding agents can resolve or dismiss; findings go stale when later pushes change the affected code
- Gates per repository for branch freshness, human approvals, agent findings, and required CI checks, with configurable bypass permissions
- CI integrations with Buildkite and Depot (public preview): builds start on push and results appear as merge gates, with Buildkite and Depot tools for converting existing GitHub Actions workflows
- Git-compatible hosting on Entire's own source control server, with repositories placed in one primary region and replicated to others for data residency and faster agent clones
- GitHub mirroring through the Entire GitHub App, with GitHub kept as the source of truth and Entire-native branches (prefix entire/unmirrored/) that accept pushes while GitHub is down
- Open-source, MIT-licensed Entire CLI that captures coding-agent sessions as Checkpoints linked to commits, with built-in support for Claude Code, Codex, Cursor, Copilot CLI, Factory Droid, OpenCode, Pi, and Antigravity CLI and a protocol for adding other agents
- CLI commands for agents and people: trail create, link and finding, search across sessions and code, checkpoint explain, session resume across agents, blame and why (labs), and a multi-agent review of a branch
- Continuous Memory: open-source entire-graph builds a local map of definitions, callers, and change impact for agents, and entire-brain turns past sessions and decisions into shared, queryable memories that appear on Trails
- Agent Insights plugin that breaks down token usage by model and activity across Trails and repositories
Limitations
- The forge launched on October 26, 2026; Trails, native hosting, and Continuous Memory are new, and CI integrations are labeled public preview
- Public repositories and self-hosted Git nodes are listed as upcoming features, so native repositories are private and there is no self-hosted option today
- CI integrations cover only Buildkite and Depot so far; GitHub Actions workflows must be converted to Buildkite or Depot CI, or kept on GitHub by using a mirror
- On GitHub mirrors, CI and merging run through a linked GitHub pull request, Entire and GitHub reviews are separate, and GitHub branch rules can still block a merge after Entire's gates pass
- Documented clone URLs for Entire-hosted repositories and mirrors use entire:// remotes, which need the git-remote-entire helper that ships with the CLI; mirrors do not support Git LFS, and the GitHub migration path does not transfer LFS content, submodules, issues, pull requests, or releases
- Review agents are billed from prepaid credits at model token rates, so the cost per push depends on model choice, diff size, and how many agents and fix loops run; agents stop when the balance hits zero
- Entire's subprocessor list shows Daytona, E2B, Anthropic, and OpenAI processing data in the US, which teams choosing EU or India hosting for data residency should factor in for review-agent runs
- No compliance certifications such as SOC 2 are listed in Entire's documentation or security pages
- Entire names Claude Code and Codex as its review-agent harnesses, and its billing page lists only Anthropic and OpenAI models for agents
Use cases
- Gating merges of agent-written changes on security, code review, and Slop scores instead of reading every line of the diff
- Letting a coding agent read a Trail's open findings through the CLI, fix them, and push until the scores stabilize
- Writing a custom review agent, such as a database migration risk check, that runs on every push to selected repositories
- Mirroring a busy GitHub monorepo into regions close to agent fleets so parallel clones stop hitting GitHub rate limits
- Hosting private repositories in a chosen region such as the EU or India for data residency
- Tracing a line of code back to the agent session, prompt, and decisions that produced it
Our take
Entire makes a clear bet: if agents write most of the code, the review step should be agents too, with people reading scores, findings, and the session history instead of every diff line. Trails put that in a usable form. Multiple review agents score each push, the last five runs show whether the fix loop has settled, and Gates turn findings, approvals, and CI into explicit merge rules. Custom agents are just instructions plus a Claude Code or Codex harness, which makes them easy to adapt to a team's own risks. The pricing is also easy to trial, since everything except agent runs is free and there is a $100 credit. The caveats are about maturity and fit. The forge launched in late October 2026, CI integrations are in preview and limited to Buildkite and Depot, there are no public repositories or self-hosting yet, and GitHub mirrors still merge through GitHub, so the cleanest experience means moving repositories to a new host. Teams that want to test the model without migrating can mirror one GitHub repository, enable the preset agents, and judge whether the scores and findings change how they merge.
Who should use it
Teams where Claude Code, Codex, Cursor, or other agents produce most pull requests and human review has become the bottleneck, teams that want custom review checks run automatically on every push, and organizations that need regional Git hosting or relief from GitHub rate limits for parallel agents.
Who should skip it
Public open-source projects, enterprises that require self-hosting or certified compliance today, teams that want to keep GitHub Actions on Entire-hosted repositories or depend on Git LFS, and developers looking for a coding agent rather than a place to host and review agent-written code.
Strengths
- Review agents, CI checks, approvals, and findings sit together on one Trail with explicit merge gates
- Custom review agents are plain instructions plus a harness and model choice, triggered on push, manually, or on a schedule
- Score history across the last five runs shows whether agent fix loops have converged
- Hosting, mirroring, Trails, and session capture are free; you pay only for review-agent model usage at provider token rates, with a $100 trial credit
- Works alongside GitHub as a mirror, so teams can try Trails without migrating
- The CLI, entire-graph, and entire-brain are MIT-licensed open source, and the CLI supports most popular coding agents
- Regional hosting with data residency choices and pushes that keep working through GitHub outages
Weaknesses
- Forge launched in October 2026, with CI integrations still in public preview
- No public repositories or self-hosting yet
- Only Buildkite and Depot for CI on Entire-hosted repositories; GitHub Actions workflows have to be converted
- Review-agent costs vary with model, diff size, and loop count, and runs stop at a zero balance
- No published compliance certifications, and agent sandboxes and model providers process data in the US
Entire pricing
Free
$0
- Git hosting, GitHub mirroring, and regional replication
- Trails, Gates, and session capture
- Open-source Entire CLI, entire-graph, and entire-brain
Agent usage credits
Provider token rates (prepaid)
- One-time $100 trial credit per billing account, no card required, expires after 90 days
- Minimum purchase $10; purchased credit expires 365 days after purchase
- Optional automatic top-up; agents stop at a zero balance
Free tier limits: Everything except review-agent runs is free. Agent runs draw on credits; each personal account or organization gets one $100 trial credit.
Note: Entire bills only for Agent usage, at the model provider's published per-million-token rates (for example Claude Opus 5.5 at $4 input and $20 output, GPT-6.1 Sol at $2 input and $10 output), and rates follow provider price changes. Each personal account and organization is billed separately, and usage is billed to the account that owns the repository. Entire's credits do not cover CI; builds run through your Buildkite or Depot account.
Technical specs
API pricing
No separate API charge is documented; Entire bills only for review-agent runs, paid from prepaid credits at the model provider's published token rates
Available models
Where Entire excels
Merging agent-written changes without line-by-line review
Security, code review, and Slop agents score each push, findings must be resolved or dismissed, and Gates require approvals and passing CI, so a reviewer can decide from the Trail rather than the raw diff.
Closing the loop between a coding agent and reviewers
A coding agent can run entire agent-help, read a Trail's open findings through the CLI, fix them, and push again until the scores stabilize, with each session recorded for the human reviewer.
Team-specific review checks
A custom agent written in plain instructions, such as a migration-risk or payments reviewer, runs on every push to the repositories it is assigned to and reports a score and findings like the presets do.
Parallel agents on a large repository
Mirroring a GitHub repository into regions near your agents serves clones and fetches from Entire, and Entire-native branches accept pushes even when GitHub is unavailable.
How Entire fits alongside other tools
Entire and Claude Code
Claude Code is a coding agent that writes and edits code; it is also one of the two harnesses Entire uses for its review agents. A common setup has Claude Code make changes and push to an Entire Trail, then read the review agents' findings through the Entire CLI and fix them, with the session saved as a Checkpoint.
Entire and OpenAI Codex
Codex writes code from your terminal, IDE, or the cloud, and is the other harness behind Entire's review agents. Entire hosts the repository, runs Codex-based or Claude Code-based reviewers on each push, and records the Codex sessions behind each commit.
Entire and GitHub Copilot
GitHub Copilot adds code completion, a cloud coding agent, and pull request review inside GitHub. Entire is a separate forge that can mirror GitHub repositories; on a mirror, Entire's Trails and review agents run alongside the GitHub pull request, and Entire and GitHub reviews are tracked separately.
Frequently asked questions
What is Entire?
Entire is a Git forge for teams that write code with AI agents. It hosts Git repositories (natively or as GitHub mirrors), replaces pull requests with Trails where review agents score each push and file findings, enforces merge Gates, and stores the agent sessions behind each commit so people and agents can see why code changed. It was co-founded by former GitHub CEO Thomas Dohmke, who is its CEO.
What is a Trail in Entire?
A Trail is Entire's replacement for the pull request. It starts from a goal, bug report, or feature request, links branches from one or more repositories, and shows eval scores from review agents at the top, along with findings, CI results, agent sessions, and human approvals. Each linked change merges once its Gates are satisfied.
How much does Entire cost?
Entire bills only for its review Agents. Hosting, mirroring, Trails, session capture, and the CLI are free. Agent runs draw on prepaid credits at the model provider's published token rates, for example Claude Opus 5.5 at $4 input and $20 output per million tokens. Each billing account gets a one-time $100 trial credit that expires after 90 days, the minimum purchase is $10, purchased credit lasts 365 days, and automatic top-up is optional.
Which models and agents power Entire's review agents?
Entire says its review agents run Claude Code and Codex in isolated cloud sandboxes from Daytona and E2B. Custom agents let you pick the harness, model, and reasoning effort. Its billing page lists Anthropic models (Claude Opus 5.5 by default, plus Opus 5, Sonnet 5.5, Sonnet 5, and Haiku 5.5) and OpenAI models (GPT-6.1 Sol by default, plus GPT-6 Astra, GPT-6 Sol, GPT-5.6 Sol, GPT-5.6 Luna, and GPT-6 Luna).
Can I keep using GitHub with Entire?
Yes. You can mirror a GitHub repository into one or more Entire regions through the Entire GitHub App and keep GitHub as the source of truth. Trails work on mirrors, but CI and merging go through a linked GitHub pull request, and GitHub branch rules still apply. You can also move fully to Entire, either by detaching a mirror into an Entire project or by copying the Git history from GitHub and pushing it to an Entire repository. Issues, pull requests, releases, and Git LFS content do not move with the Git data.
Is Entire open source or self-hostable?
The forge itself is a hosted service and is not open source. The Entire CLI, entire-graph, and entire-brain are open source under the MIT license. Self-hosting of Git nodes is listed as an upcoming feature, not something available today.
Does Entire write code for me?
No. Entire hosts and reviews code and records agent sessions. You still use a coding agent such as Claude Code, Codex, or Cursor to write the changes; Entire's agents review each push, and your coding agent can read the findings through the Entire CLI and fix them.
Where can Entire host my code?
Entire's Hosting Regions page (effective October 19, 2026) lists the European Union (Frankfurt), United States (Ohio), Australia (Sydney), India (Mumbai), United Kingdom (London), and Singapore. Its launch announcement named the US, EU, India, and Australia, and some docs pages also mention Brazil, so check the region list when you create a project. You choose a primary region per project and can replicate repositories to other regions. Review-agent sandboxes and the Anthropic and OpenAI model APIs process data in the US, according to Entire's subprocessor list.
Integrations & fit
About Entire
Entire is the company co-founded and led by former GitHub CEO Thomas Dohmke. It shipped an open-source CLI in February 2026 to save coding-agent sessions (prompts, transcripts, tool calls, token usage) as Checkpoints linked to Git commits, then added a distributed Git hosting network for mirroring GitHub repositories. On October 26, 2026 it launched the full forge. The core idea is the Trail, Entire's replacement for the pull request. A Trail can start as a bug report or feature request, or from your agent, and can group branches from several repositories. When code is pushed, the review agents you have enabled run independently in isolated cloud sandboxes and each returns a score with its rationale, plus findings with severity and suggested fixes. Presets cover code review, security review, and Slop (sloppy patterns and drift), and you can write custom agents by choosing a harness (Claude Code or Codex), a model, reasoning effort, a timeout, instructions, and a trigger. Each eval shows the current and last five runs so you can see whether repeated agent fix-and-push loops have stabilized. Gates then make the merge policy explicit: branch freshness, human approvals, open findings, and CI checks, either from the Buildkite and Depot integrations (in public preview) or, on GitHub mirrors, from GitHub. Underneath is Entire's own Git-compatible server. You can host repositories natively in regions you choose (Entire's Hosting Regions page lists the US, EU, UK, India, Australia, and Singapore) or mirror a GitHub repository and keep GitHub as the source of truth, in which case merging and CI go through a linked GitHub pull request. Continuous Memory adds two open-source CLI plugins: entire-graph, a local map of your code for agents, and entire-brain, which indexes past agent sessions into memories other agents can query. Everything except the review agents is free; agent runs are paid from prepaid credits at the model provider's published token rates, with a one-time $100 trial credit. The tradeoffs are maturity and scope. The forge only launched in late October 2026, public repositories and self-hosted Git nodes are still on the roadmap, the documented clone URLs use Entire's own entire:// Git remote helper, mirrors do not support Git LFS, and no compliance certifications are published. Entire does not write features for you; it hosts, reviews, and remembers the work your coding agents do.
Updates from Entire
Entire launched its Git forge on October 26, 2026: Trails with review agents and Gates replace pull requests, repositories can be hosted natively in chosen regions or migrated from GitHub, entire-graph and entire-brain add Continuous Memory, and Buildkite and Depot CI integrations arrived in public preview.
Entire added Google sign-in alongside GitHub, so an account no longer requires a GitHub login.
Entire introduced its distributed, Git-compatible hosting network for mirroring GitHub repositories into regional clusters so agents can clone quickly without hitting origin rate limits.
Entire announced the company with a $60 million seed round led by Felicis and shipped its first product, an open-source CLI that saves coding-agent sessions as Checkpoints alongside Git commits.
Are you the founder? Claim this listing →