Getting started with CodeMind
A local Python RAG assistant for asking questions about local files and web pages.
Ask questions over local project files
CodeMind is a Python learning project that builds a local retrieval-augmented generation (RAG) assistant. Give it a web URL or a local text/Markdown file, then ask questions grounded in what it ingested.
The documented pipeline extracts text, splits it into overlapping chunks, stores local embeddings in Chroma, retrieves the top matching chunks, and gives that context to a local language model. Responses can include source context rather than relying only on model memory.
CodeMind supports two model backends: Ollama for models already installed on the machine, or llama.cpp for GGUF models managed in the project's models directory. The CLI asks which backend to use when it starts.
Inference and the vector database run locally. URL ingestion still requires a network connection to fetch the URL, and the first run may download Python packages, an embedding model, or a language model depending on the selected backend.