What Checkpoints Does
Checkpoints tackles the provenance gap that grows wider as AI agents generate more code faster than humans can review it. Traditional Git records what changed in a diff but provides no record of the prompts, reasoning steps, or tool calls that produced those changes. When agents write hundreds of lines per session, the context behind decisions evaporates once the terminal window closes. Checkpoints makes that context permanent by treating agent reasoning as first-class versioned data stored directly in the Git repository.
Installation and Metadata Architecture
The installation experience is remarkably simple for a tool addressing such a complex problem. A single curl command installs the CLI, and running entire enable in a repository configures the necessary Git hooks. The CLI detects which AI coding agent is active and configures accordingly, supporting Claude Code, Gemini CLI, OpenCode, Cursor, Factory AI Droid, and GitHub Copilot CLI. The two-step setup avoids the configuration sprawl that makes many developer tools frustrating to adopt across teams and repositories.
The metadata storage architecture is elegantly designed to avoid polluting the primary branch history. All session data lives on a dedicated branch called entire/checkpoints/v1, keeping the main branch clean for standard code review workflows. A 12-character hex checkpoint ID is added as a trailer in commit messages, creating a bidirectional link between the code change and its reasoning context. This design means standard Git operations like clone, pull, and merge work exactly as before with zero interference.
Rewind Capability and Worktree Support
The rewind capability addresses a real pain point in agent-assisted development. When an agent takes a wrong turn and generates code that breaks the project, developers can run entire rewind to restore files to any previous checkpoint state without altering commit history. The session context is preserved so the agent can resume from the known-good state with full awareness of prior decisions. This non-destructive restoration approach is safer than git reset or manual reverts that lose context.
Git worktree support enables independent session tracking across parallel workspaces, meaning developers running multiple agents on different features get isolated checkpoint histories for each. If two sessions run on the same commit, Entire warns the developer and tracks them separately with independent rewind targets. This parallel session awareness is critical for the increasingly common workflow of dispatching multiple AI agents simultaneously.
Web Dashboard and Team Background
The web dashboard at entire.io provides a visual interface for browsing checkpoint history by branch, drilling into individual sessions, and inspecting side-by-side diffs paired with the agent transcripts that produced them. Signing in with GitHub synchronizes checkpoint data to a Supabase-backed display layer, though the company explicitly states this data is used only for platform functionality and not for training or other purposes. The dashboard transforms code review from examining diffs in isolation to understanding the full decision chain.
The pedigree behind Checkpoints gives it unusual credibility in developer tools, but the funding context should be treated as company background rather than proof of CLI capability. Thomas Dohmke served as GitHub CEO from 2021 through August 2025 and helped scale GitHub Copilot. Launch and funding posts around Entire described a $60M seed round at a $300M valuation with backers such as Felicis, Madrona, M12, Garry Tan, and Olivier Pomel; the official product and docs surfaces focus on Checkpoints' Git-native session capture, audit trail, and broader platform roadmap.
Platform Vision and Limitations
The broader Entire platform vision extends well beyond checkpoint capture. The company plans three components: a Git-compatible database unifying code, intent, constraints, and reasoning; a universal semantic reasoning layer enabling multi-agent coordination; and an AI-native software development lifecycle for agent-to-human collaboration. The current CLI serves as the data collection layer for this larger vision, meaning early adoption captures value immediately while positioning teams for future platform capabilities.
Limitations at this stage are worth noting. The CLI is relatively new, launching in February 2026, so edge cases with large repositories and high-frequency agent sessions are still being discovered. Community discussion on Hacker News raised valid concerns about metadata storage growing large when agents produce megabytes of context per session. The web dashboard is under active development with limited visualization capabilities compared to the long-term roadmap. Some developers question whether simpler Git hooks could achieve similar checkpoint results without a dedicated tool.
The Bottom Line
Checkpoints solves a genuine problem that becomes more pressing as agent-generated code volume increases. The ability to trace any change back to the reasoning that produced it is valuable for code review, onboarding, compliance, and debugging. The clean Git-native implementation and simple two-step setup lower the barrier to adoption, while the Entire platform roadmap signals long-term investment in making agent-human collaboration auditable and reproducible.