Quivr

Quivr is an open-source RAG (retrieval-augmented generation) and knowledge-base toolkit for developers, enabling quick integration of files, vector backends and LLMs to build assistant-style applications.

Key features

• Opinionated RAG workflows: default but customizable retrieve-and-generate pipelines with reranking strategies.
• Wide model and vector store support: compatible with OpenAI, Anthropic, Mistral and vector backends like PGVector and FAISS.
• Lightweight deployment and examples: quivr-core library, notebooks and quickstart scripts for both self-hosting and cloud usage.

Use cases

• Document QA and knowledge assistants: expose product docs or internal knowledge as a conversational interface.
• Interactive analysis: turn arbitrary files into queryable knowledge and compose complex retrieval tasks.
• Prototyping: rapidly spin up demos to evaluate retrieval/generation strategies and model combinations.

Technical notes

• Core library & SDK: quivr-core provides a simple API to build a “Brain” from files and query it interactively.
• Configurable workflows: define retrieval and generation nodes with YAML configuration, with support for tools and web search.
• Community and ecosystem: extensive examples, plugins and documentation at core.quivr.com with an active contributor base.