AI agents are churning out hundreds of commits a day. But ask "why was this code written this way?" and nobody can answer. The prompts are gone, the reasoning is gone, the decision-making context is gone. The former GitHub CEO just bet $60M on this problem.
What is this?
Entire is a platform that records the "why" behind AI agent-generated code. On February 10, 2026, Thomas Dohmke (former GitHub CEO) launched it alongside a $60M seed round. At a $300M valuation, it's the largest seed round ever for a developer tool.
Dohmke served as GitHub CEO for 4 years starting in 2021, scaling up Copilot during that time. When he left GitHub in August 2025, he said: "After 10+ years at Microsoft, 7 at GitHub, and 4 as CEO, I felt the urge to be a founder again."
The problem he identified is clear. AI agents are mass-producing code, but traditional Git only records "what changed," not "why it changed." Prompts, reasoning processes, decision-making context — all lost. In Dohmke's words: "Code is being produced in massive volumes, far faster than any human can reasonably understand."
Entire's first product is Checkpoints. It's an open-source CLI tool that records entire AI agent coding sessions — the original prompt, the agent's reasoning steps, implementation logic, and decision patterns. This metadata is stored alongside Git commits, so later when someone asks "why is this code like this?" there's actually an answer.
What's technically interesting is the storage approach. Checkpoints doesn't touch the main branch. It stores session data in a separate branch called entire/checkpoints/v1, keeping the existing commit history clean. Fully compatible with Git, yet solving the "context preservation" problem Git can't handle.
It currently supports 5 agents: Claude Code, Gemini CLI, OpenCode, Cursor, and Copilot CLI. Felicis Ventures led the round, with participation from Microsoft M12, Madrona, Basis Set, and angel investors including Datadog CEO Olivier Pomel, Y Combinator CEO Garry Tan, and Jerry Yang (former Yahoo CEO).
Why this matters
When the EU AI Act goes into full effect in August 2026, audit trails for high-risk AI systems become a legal requirement. Recording the decision-making process behind AI-written code isn't just a nice-to-have — it's about to become a compliance necessity.
What changes?
Honestly, Git alone handles code management fine. But in the age of AI agents, it's not enough. Let's compare.
| Traditional Git Workflow | Entire Checkpoints | |
|---|---|---|
| What's recorded | Code changes (diffs) only | Prompts + reasoning process + decisions + code changes |
| Context preservation | Relies on commit messages (human-written) | Automatic capture of entire AI session |
| Code review | "Why was this written this way?" → Have to ask the author | Instantly check reasoning process in session logs |
| Repeated mistakes | Same agent can repeat the same mistakes | Can reference past troubleshooting records |
| Audit trail | Code-level only | Full traceability from prompt to result |
| Cost management | No token usage tracking | Records token usage and third-party tools used |
Dohmke made a key point: "In 2026, every leader needs to think about labor costs not just in terms of salaries, benefits, and travel, but also in tokens." Entire records token usage too, making AI agent operating costs transparent.
It also solves the PR review bottleneck. Dohmke said: "After writing code, the next thing you do is code review. But pull requests have the same problem." When AI creates hundreds of PRs a day, understanding the context of each one hits limits with traditional methods. Since Checkpoints leaves session logs, reviewers can instantly grasp "why was this changed?"
Felicis founder Aydin Senkut put it this way: "Forcing agents into human-centric workflows creates friction and bottlenecks. You need to rethink the platform itself for agent-first environments."
The essentials: how to get started
Checkpoints is open source and ready to use right now.
- Install it
Homebrew is the easiest way.brew tap entireio/tap && brew install entireio/tap/entire
Go users can also rungo install github.com/entireio/cli/cmd/entire@latest. - Enable it in your project
Inside a Git repo, runentire enable --agent claude-code. Choose from Claude Code, Gemini CLI, OpenCode, Cursor, or Copilot CLI. If you use multiple agents, enable each one separately. - Code as usual
Code with your AI agent and git commit — Checkpoints automatically captures session metadata. No extra steps needed. Your main branch stays clean. - Review sessions
Useentire statusto see current session info, andentire explainto get an AI summary of a session or commit. You can also browse visually on the web UI (entire.io). - Rewind (when needed)
If something went wrong, useentire rewindto restore to a previous checkpoint. It's non-destructive and doesn't touch Git history, so it's safe.
Tip: Store checkpoints in a separate repo
If you want to keep session data private in a public repo, use entire enable --checkpoint-remote github:myorg/checkpoints-private to store it in a separate private repo. Secrets are automatically redacted, but it's best-effort, so this approach is recommended for sensitive projects.
