ats4321/ragit

ragit is a command-line tool that lets you have a conversation with a folder of documents on your own computer. You point it at a folder containing text files, Markdown files, PDFs, or Word documents, and then ask questions in plain language. It finds the relevant parts of those documents and uses a local AI model to write an answer, showing you which document sections it pulled from. The tool uses a technique called Retrieval-Augmented Generation (RAG). When you run the index command, ragit reads all supported files in a folder, breaks them into overlapping chunks of roughly 500 words each, and converts those chunks into numerical vectors using a local model called nomic-embed-text. Those vectors are stored in a local database called ChromaDB, which keeps everything on your machine. When you ask a question during chat, ragit converts your question into a vector, finds the most similar document chunks, and feeds them to a local chat model to produce an answer. Everything runs locally. No API key is needed, no files leave your machine, and no internet connection is required after the initial model downloads. The default chat model is llama3.2, but any model installed in Ollama will work. Both Ollama (the tool that runs local AI models) and Python 3.10 to 3.13 are required. The three main commands are concise: ragit index <folder> scans and indexes documents, ragit chat <folder> starts the conversation, and ragit clear <folder> removes the stored index. A ragit models command lists available Ollama models. The README is brief and does not cover performance, folder size limits, or configuration options beyond these basics.