CodeTalk – recover why AI-written code was written, quoted from Git (zero-LLM) (github.com)

CodeTalk has been introduced as a tool designed to address a growing concern among developers regarding trust in AI-generated code. By leveraging existing commit history in Git, CodeTalk provides real citations of code changes, allowing users to understand the reasoning behind specific coding decisions. This is significant as it directly counters the prevalent distrust in AI outputs; a Stack Overflow survey revealed that nearly half of developers lack trust in AI-generated code, primarily due to experiences of fabricated "explanations" from AI models. The tool operates without reliance on large language models (LLMs), maintaining a 'zero-LLM' approach for local data retrieval and decision tracking. It uses SHA verification to link code changes with actual commit messages, including optional decision notes (Vibe-Decision/Vibe-Watch) that help ground the coding history. CodeTalk's architecture focuses on privacy and data integrity by processing information on the user's machine without cloud dependency, thereby enhancing transparency in the coding process and enabling developers to engage in verifiable discussions about their code's evolution.