Re_gent

Product Introduction

1. Definition: Re_gent is a specialized version control system (VCS) for AI agent activity. It is a local-first CLI tool that acts as an audit and control layer for autonomous coding agents.
2. Core Value Proposition: It exists to provide developers with visibility, traceability, and rollback capabilities for changes made by AI coding assistants. It solves the "black box" problem of AI-driven development by tracking the link between prompts and code modifications, enabling safe and controlled agentic workflows.

Main Features

1. rgt log - AI Agent Session Logging: This command provides a detailed, chronological audit trail of all agent actions. Unlike git log which shows file commits, rgt log displays the full conversation history, including the exact prompts that led to specific code changes. It works by intercepting and content-addressing every interaction between the developer's IDE and the AI agent's API, storing prompts, responses, and file diffs in a local, immutable database.
2. rgt blame - Prompt-to-Code Lineage: This feature enables granular traceability, mapping every line of code in a file back to the specific AI agent prompt that generated or modified it. It works by maintaining a fine-grained, content-addressed graph that links code blocks to their generative origin, even across multiple agent sessions or branches, providing a complete audit trail for debugging and compliance.
3. rgt sessions - Parallel Agent Branching: This functionality allows multiple AI agent conversations (or "sessions") to operate in parallel on the same codebase without conflict. Each session is isolated in its own branch. It works by leveraging a git-like branching model under the hood, where each agent's conversational context and proposed changes are contained separately, allowing developers to review and merge agent work selectively.

Problems Solved

1. Pain Point: The loss of context and auditability in AI-assisted coding. Traditional git cannot answer why a change was made by an AI, which prompt caused a bug, or how to revert an agent's entire conversational thread without losing valuable work.
2. Target Audience: Software engineers and development teams using AI coding agents (e.g., with Claude Code, Cursor, Continue.dev); engineering managers requiring oversight and quality control for AI-generated code; and developers practicing "agentic workflows" where multiple autonomous or semi-autonomous AI agents collaborate.
3. Use Cases: Debugging a regression introduced by an AI refactoring agent; auditing code quality and reasoning behind AI-suggested changes; safely experimenting with different AI agents or prompts on the same feature; and maintaining a reproducible history of AI-assisted development for compliance or knowledge retention.

Unique Advantages

1. Differentiation: Unlike git, which tracks file states, Re_gent tracks the agent conversations and actions that produce those states. Unlike simple chat history in an IDE, Re_gent provides structured version control operations (log, blame, checkout, branch) specifically designed for the non-linear, prompt-driven nature of AI coding.
2. Key Innovation: Its core innovation is the application of content-addressed storage and git-like principles to the stream-of-consciousness output of AI agents. By immutably storing every prompt, completion, and resulting file diff, it creates a replayable, branchable timeline of AI activity that persists even if the upstream AI tool (like Claude) compacts or deletes its conversation history.

Frequently Asked Questions (FAQ)

1. How does Re_gent differ from my IDE's chat history? Your IDE's chat history is ephemeral, linear, and often lost on refresh or compaction. Re_gent is a persistent, queryable version control system that branches sessions, links code diffs directly to prompts, and allows you to surgically undo or replay specific agent actions across your entire codebase.
2. Can I use Re_gent with GitHub Copilot or ChatGPT? Currently, Re_gent's public alpha integrates with Claude Code and similar agentic platforms. Support for other tools like Cursor, Continue.dev, and broader AI coding assistants is on the roadmap, as the tool is designed to be platform-agnostic by intercepting agent activity.
3. Is my code or prompt data sent to a remote server? No. Re_gent is architected as a local-first CLI tool. All audit data—prompts, code diffs, and session history—is stored locally on your machine in a content-addressed database, ensuring privacy and full ownership of your intellectual property and AI interaction data.
4. What happens if my AI agent uses the /compact command? This is a key problem Re_gent solves. When an AI agent compacts its context, it destroys its own operational history. However, Re_gent has already captured and stored the full, uncompacted conversation and all associated actions locally. Your audit trail remains intact even after the agent forgets.
5. Do I need to change my git workflow to use Re_gent? No. Re_gent operates as a complementary layer alongside git. You use git for managing collaborative human commits and rgt for managing and understanding autonomous AI agent activity. They are designed to work in tandem for modern, hybrid human-AI development teams.