Entire launches a Git forge with Trails replacing pull requests

• Entire’s Git forge is live for agent-human collaboration, as announced in the launch thread.
• Trails replace conventional PRs with agent-driven review loops, the change Entire investor GergelyOrosz highlighted at launch.
• Review agents run in isolated cloud sandboxes, with E2B confirming its involvement.
• Hosting and memory are free; AI review consumes prepaid credits, according to the billing explanation.

A single Trail can span multiple repositories. GitHub mirrors also have an entire/unmirrored/ branch escape hatch for agent pushes during an upstream outage.

Trails’ most interesting addition is a six-run eval history: the current run plus five previous runs. Entire’s launch post describes three stages:

1. Start the work. Open a Trail from a bug report, feature request, or coding agent, before or after implementation begins.
2. Score each push. Independent review agents inspect the commit and return scores with rationales. Entire’s own repositories use evaluators including Security, Slop, and Human Review Required.
3. Clear findings and merge. Humans or agents can address or dismiss findings. Repository-level Gates can require findings to be cleared, human approval, and passing CI checks before approval.

Open findings stay visible; resolved findings move out of the main view but remain retrievable. The dashboard exposes whether successive coding and review loops are stabilizing, which is the stated purpose of the score history.

Entire runs Claude Code and Codex inside isolated cloud sandboxes supplied by Daytona and E2B, according to the announcement. Review agents can be configured through the web interface or repository .json files.

LangChain’s OpenSWE also reviews every PR, according to an account of the company’s internal agents; its agent process stays outside the sandbox used to execute code.

Entire’s Git-compatible server combines a distributed object store with a transactional ref database. Its published, first-party benchmark measures push throughput:

| Forge | Pushes per second |
| --- | ---: |
| Entire | 418 |
| GitHub | 4.7 |
| Cursor Origin | 65 |
| Pierre | 15 |
| GitLab | 10 |

The advertised “up to 89x” advantage over GitHub refers to these throughput results.

Hosting supports:

• Regional placement: US, EU, Australia, and India at launch, with a primary region and optional replication.
• Native repositories: Created through the web interface or Entire CLI.
• GitHub mirrors: GitHub can remain the source of truth, with Trails layered over existing PRs. Push traffic still routes through GitHub.
• Migration: Fetch-and-push, the entire-migrate plugin, or detaching an existing US-region mirror.

Entire’s Continuous Memory connects recorded coding sessions to a map of the repository through two open-source, local-first plugins:

• entire-graph: Builds a code map with definitions, callers, types, routes, change impact, and file:line locations. Its built-in tree-sitter analyzer makes no network requests or model calls and requires no API key.
• entire-brain: Indexes that map alongside past agent sessions recorded through the Entire CLI, preserving decisions and previous attempts for retrieval in later sessions.

Connecting entire-brain to Entire’s backend makes memories shared across the team. Trails can then display the memories created during the work, exposing decisions alongside review evidence.

Buildkite and Depot are the initial CI integrations. Entire configures CI at the project level, allowing repositories in a project to share infrastructure, according to the launch announcement.

• Buildkite: bk pipeline convert converts GitHub Actions workflows. Builds can run on Buildkite’s infrastructure or the team’s own infrastructure.
• Depot: depot ci migrate discovers GitHub Actions workflows, promotes them to Depot CI workflows, and flags items needing review. Companion commands import secrets and variables.

Entire paired the launch with a developer walkthrough.

The billing documentation specifies prepaid model-token usage rather than charges for hosting or memory:

• Trial: $100 per billing account, no card required, expiring after 90 days.
• Exhausted balance: Agents stop across every region for that billing account; other features remain available.
• Automatic top-up: Off by default. Its default configuration replenishes the balance to $50 below a $10 threshold, with no monthly maximum.
• Purchased credits: Minimum purchase of $10, expiring after 365 days.

Entire investor GergelyOrosz described token margins and future enterprise features as the intended funding model. The billing page currently says Agent usage is charged at model providers’ published rates.

Usage reporting can lag by several minutes, and in-progress model calls finish. A small overrun beyond a zero balance can therefore appear on the monthly invoice or be covered by the next top-up.