GitHub's re_gent brings Git-style version control to AI coding agents, solving the problem of untracked agent activity. The tool automatically captures every Claude Code interaction, creating an immutable audit trail of code changes, prompts, and tool calls. Unlike traditional Git, re_gent tracks agent-specific workflows, allowing developers to blame individual lines of code to their exact prompts and rewind sessions when errors occur.

The system stores agent activity in a .regent/ directory using BLAKE3 hashing for deduplication and SQLite for fast queries. Each step in the DAG contains workspace snapshots, conversation transcripts, and tool arguments. Users can list active sessions, compare changes with `rgt log`, and inspect specific steps with `rgt show`. The tool integrates seamlessly with existing Git workflows, creating a dual-layer version control system for both code and agent behavior.

Installation via Homebrew or Go install is straightforward, with optional VS Code extension support for inline blame annotations. The project emphasizes conversation tracking, surviving session restarts and `/compact` operations that typically erase chat history. Early adopters report improved debugging efficiency, though the repository warns it's still POC-level with active development.

While complementary to Git, re_gent addresses unique challenges in AI development: tracking agent decision-making processes and maintaining context across interrupted sessions. The open-source release invites contributions to expand its capabilities beyond Claude Code integration. As AI agents become more autonomous, tools like re_gent may become essential for maintaining codebase integrity and accountability.