Developer Tools

Show HN: Git for AI Agents

Automatically tracks every tool call and links changes to prompts.

Deep Dive

Regent introduced re_gent, a Git-like version control system designed specifically for AI agent activity. The tool automatically captures every tool call an AI agent makes — edits, file writes, bash commands — without requiring any manual commits. Users can then run `rgt log` to see a chronological history of what the agent did, including the exact file changes and the context of the conversation that triggered them. The `rgt blame` command traces any line in the codebase back to the specific prompt that caused it to be written, providing full auditability. A `rgt rewind` feature is in development to restore the workspace to any previous step.

Technically, re_gent stores agent activity in a `.regent/` directory using BLAKE3 content-addressed blobs and an SQLite index for querying. Each tool call creates a "Step" with a parent hash, workspace snapshot, transcript delta, and session ID. Steps form a DAG, with each agent session maintaining its own branch. Installation is via Homebrew (`brew install regent`) or Go (`go install`), and a VS Code extension provides inline blame annotations. The tool supports tracking multiple concurrent agent sessions, making it ideal for teams using Claude Code or similar AI coding assistants in production environments.

Key Points
  • Automatically captures every AI agent tool call (edit, write, bash) without manual commits.
  • `rgt blame` traces any line of code back to the exact prompt that generated it.
  • Supports multiple concurrent sessions with a DAG-based step history and SQLite index.

Why It Matters

Makes AI agent actions auditable and revertible, critical for production codebases where debugging is essential.