Quivr

Why we included this project

Quivr is a Python framework that handles the retrieval-augmented generation plumbing so you don't have to build it from scratch. Install quivr-core, point a Brain at a folder of files, and you can ask questions over that content in a few lines of code. It keeps the two pieces that usually cause lock-in swappable: the LLM (OpenAI, Anthropic, Mistral, or local models through Ollama) and the vector store (PGVector, Faiss, and others). The retrieval pipeline itself is configurable through a YAML file, so you can add a reranker, change history handling, or enable internet search and tools without touching application code. Teams that want a working RAG baseline they can extend, and that care about swapping models and stores later, will appreciate the opinionated defaults.