quivr

Quivr is an open-source RAG (Retrieval-Augmented Generation) framework designed to help developers integrate generative AI capabilities into their applications. Marketed as a 'second brain' empowered by Generative AI, it allows users to ingest files and query them using large language models. Key features include: - **Opinionated RAG**: A pre-configured, efficient RAG pipeline so developers can focus on their product rather than infrastructure. - **LLM Flexibility**: Compatible with various LLMs including OpenAI, Anthropic, Mistral, and local models via Ollama. - **File Support**: Works with multiple file formats such as PDF, TXT, and Markdown, with the ability to add custom parsers. - **Customization**: Allows customization of the RAG workflow, including adding internet search and tools. - **Megaparse Integration**: Integrates with Megaparse for file ingestion. The framework can be installed via pip (`pip install quivr-core`) and requires Python 3.10 or newer. Developers can create a functional RAG 'Brain' with just a few lines of code. Quivr also supports YAML-based configuration for defining RAG workflows, including settings for history filtering, retrieval, reranking (e.g., using Cohere), and LLM parameters like temperature and token limits. The project is licensed under the Apache 2.0 License.