re_gent
A Git-like version control for AI coding agents that records prompts and tool calls, helping developers trace exactly which agent session and instruction produced each line of code.
Tool overview
re_gent captures the full step-by-step journey of AI coding agents. Instead of only showing what changed in code, it logs the agent's tool usage, file modifications, and the prompts that triggered each action. Commands such as rgt log reveal session timelines, rgt blame maps individual lines back to a specific prompt and session, and rgt revert rewinds both code and conversation state, maintaining history even after sessions are compacted or fragmented.
The primary benefit is an unprecedented audit trail for agent-assisted development. Git tells you who changed a line, but re_gent answers the critical question: which prompt or tool call caused this code? This makes debugging complex agent behaviors, reviewing team agent interactions, and rolling back misguided steps far less error-prone. It transforms agent output from a black box into a traceable, controllable process.
Potential downsides and risks: As a nascent open-source project, its sample evidence is limited, and it has not yet demonstrated stability at enterprise scale. It supplements rather than replaces Git, meaning developers must manage two layers of history.