Introducing GAIT: Git for Artificial Intelligence Tracking

We version code with Git — but AI conversations are still ephemeral. And AI conversations, our prompts, context, and LLM responses; need *context* control; like version and source control before it. Once you close the terminal, the reasoning path, exploration, and decisions are gone. In this video, I introduce GAIT, a version control system for AI conversations, and gaithub, a remote that lets you push, clone, and resume AI conversations anywhere. GAIT treats conversations as first-class artifacts: Every turn is a content-addressed object Every conversation state is a commit Branches represent alternate reasoning paths Merges represent choosing the best answer Memory is explicit and intentional Tokens are visible and accountable In this demo, I show: Initializing a GAIT repo Chatting with a local LLM (Ollama) Automatic commits per turn Resuming conversations deterministically Branching to explore different explanations Merging the best reasoning path Pinning memory and inspecting token budgets Pushing the conversation to a remote (gaithub) Cloning the repo elsewhere and resuming the conversation Pushing changes back — distributed cognition in action This is not prompt logging. This is not chat history export. This is version control for thinking. Roadmap Web UI to browse conversations in the browser gaithub.com with real DNS, users, and tokens Forks and pull requests — but for reasoning Community-driven, open AI conversations Model experimentation via branches (local and cloud) Git changed how we build software. GAIT is about changing how we build thinking.