Table of Contents
Quick Summary
- Entire is a developer platform designed to act as an observability and audit layer for AI-generated code.
- Launched by former GitHub CEO Thomas Dohmke, it introduces Checkpoints, a new Git snapshot mechanism that versions AI reasoning traces, tool calls, and transcripts alongside your commits.
- By shunting this metadata to a dedicated hidden branch (
entire/checkpoints/v1), Entire keeps your primary history clean while providing a "rewind" button for your repository and a searchable record of the "why" behind every AI-assisted change.
What is Entire?
Entire is a developer platform designed to record, version, and manage AI coding sessions directly within your Git workflow. It serves as an observability and audit layer for code generated by AI agents like Claude Code and Gemini CLI.
The project was launched by former GitHub CEO Thomas Dohmke and recently raised a $60 million seed round to address the challenges of managing agentic code at scale.
How Entire CLI Works: The Observability Layer for AI-Generated Code
Entire introduces Checkpoints, a new Git primitive that automatically versions agent context alongside your code.
- Metadata Integration: It captures the full interaction, including transcripts, prompts, tool calls, and token usage.
- Shadow Branches: During active sessions, Entire uses temporary shadow branches to hold snapshots of your code. This allows you to rewind your repository to any previous state if an agent makes a mistake.
- Clean History: To keep your primary Git history uncluttered, Entire stores all permanent session metadata on a dedicated, hidden branch named
entire/checkpoints/v1. - Bidirectional Linking: It adds a 12-character Checkpoint ID to your Git commit messages as a trailer, linking your code changes directly to the reasoning on the entire.io dashboard.
The Key Insight: Closing the "Provenance Gap"
The platform solves the "provenance gap" inherent in AI-assisted development. While traditional Git tells you what changed, the reasoning behind those changes often evaporates once an AI's context window closes.
By treating AI reasoning as a first-class, versioned primitive, Entire makes the "thought process" behind a change searchable and shareable. This allows human reviewers to understand the context of AI-generated code during pull requests, leading to more informed feedback.
Advanced Features
- Nested Session Support: When a primary agent spawns sub-agents (e.g., using Claude Code’s Task tool), Entire captures these as a hierarchy, preserving the full context of multi-agent workflows.
- Line Attribution: Entire calculates the exact percentage of code written by the agent versus the human developer by tracking changes before and after each agent run.
- Explain Commits: You can use the
entire explain --generatecommand to produce AI-powered summaries of specific sessions, highlighting intent, outcomes, and friction points.
Getting Started with the Entire CLI
1. Prerequisites
- Git installed and configured.
- Claude Code or Gemini CLI (preview) installed and authenticated.
- OS: Linux, macOS, or Windows with WSL.
2. Installation
Using Curl (Linux/WSL):
curl -fsSL https://entire.io/install.sh | bash
Using Go:
go install github.com/entireio/cli/cmd/entire@latest
With Homebrew (macOS):
brew tap entireio/tap brew install entireio/tap/entire
You can verify if the CLI is installed correctly using command:
entire version
3. Project Setup
Navigate to your project repository and run:
entire enable
This installs the necessary hooks. By default, it uses the manual-commit strategy, meaning checkpoints are only permanently archived when you make a Git commit, keeping your history clean.
4. Typical Workflow
- Start Agent: Run your assistant (e.g.,
claudeorgemini). - Work Normally: Entire tracks changes on shadow branches in the background.
- Check Status: Use
entire statusin a separate terminal to see active session details. - Rewind: If the AI makes a mistake, run
entire rewindto restore your code to a previous save point. - Commit & Push: When finished, commit and push your changes. Entire will prompt you to link the session, pushing the metadata to the
entire/checkpoints/v1branch for cloud visibility.
5. Using the Web Dashboard
Visit entire.io to sign in with GitHub and authorize the Entire App. Once synced, you can browse your activity feed, view side-by-side diffs, and inspect full AI transcripts to understand the "why" behind every change.
Addressing Common Misconceptions
The developer community has debated whether Entire is "premature infrastructure," but the platform offers capabilities that basic scripts cannot easily replicate:
Misconception 1: "It's just Git hooks"
While Entire uses hooks, it performs complex state management via shadow branches to enable non-destructive rewinding and handles the recursive complexity of nested sub-agents.
Misconception 2: "It bloats the repository"
All heavy metadata (JSON transcripts) lives on the separate entire/checkpoints/v1 branch. Your main branch only carries a small 12-character ID in the commit trailer.
Misconception 3: "Agents already have history"
Local histories (like Claude's /resume) are siloed on a single machine. Entire enables collaborative observability, allowing team members to see an agent's logic during code reviews on the web dashboard.
What's Your Thoughts on Entire CLI?
As stated earlier, Entire CLI has sparked significant debate in the developer community. Some see it as a necessary "observability layer" for the era of agent-generated code, while others question if the same functionality could be achieved with simple custom scripts or Git hooks.
What do you think of this project? Is it really necessary? Please share your thoughts in the comment section below.
Resources:
