thegaf/gafbrain

GafBrain is a tool that takes your exported ChatGPT conversation history and turns it into a searchable personal knowledge library, or Brain, that you can then hand to any AI assistant, including ChatGPT, Claude, or Gemini. The idea is that your original conversation exports stay on your own computer, and only the compiled Brain gets shared with whichever AI you choose to use, so you are not locked into one AI provider to keep access to your own accumulated history. Right now the project is at an early, alpha stage and only fully supports ChatGPT export files. Support for Reddit, Claude, and Google Drive is listed as in progress, and the README lists Gmail, Photos, GitHub, LinkedIn, documents, presentations, and code as planned for the future but not yet built. To use it, you install Node.js, download the project, run its setup command to create a private workspace folder on your machine, and then copy your exported ChatGPT conversation files into a specific folder inside that workspace. Running the build command processes those files: it finds the exports, cleans up the conversation data, builds a search index, creates a timeline of when conversations happened, and packages everything into a bundle meant to be uploaded to an AI assistant. A verify command checks that each of these steps completed successfully. Once built, you point an AI assistant at the resulting Brain folder and give it a short instruction telling it to treat that folder as its source of truth and to say when it cannot find an answer rather than guessing. From there you can ask questions like what a particular past conversation was about, when you first mentioned a certain topic, or what themes come up repeatedly across your history. When a newer ChatGPT export becomes available, you move the old export into an archive folder, copy in the new one, and rebuild. The project separates the reusable compiler code from your personal workspace folder, so updating GafBrain to a newer version does not touch or overwrite your existing personal data. It is written in JavaScript, runs entirely locally through npm commands, and is released under the MIT license.