Introducing Quivr, an Open-Source RAG Framework

It's a framework focused on automating the complex plumbing involved in implementing RAG (Retrieval-Augmented Generation), such as file parsing, embedding, vector store connections, and conversation history management.

Key Features and Design Principles:

• Model and Storage Independence: It broadly supports commercial APIs like OpenAI, Anthropic, and Mistral, as well as local models via Ollama, and is not tied to any specific vector database.
• Flexible Data Input: It supports various file formats such as PDF, TXT, and Markdown, and allows adding custom parsers.
• Configuration-Centric Pipeline: Steps like retrieval, reranking, and LLM calls are described as nodes and edges in a YAML configuration file rather than in code, making it easy to change the structure.
• Improved Productivity: The core package, quivr-core, allows RAG functionality to be implemented with minimal code, helping developers focus on actual product logic rather than building infrastructure.