Quivr is an open-source retrieval-augmented generation framework originally presented as a generative-AI “second brain.” Its core package focuses on giving developers an opinionated RAG pipeline that can ingest personal or application documents and answer questions over them without requiring every project to rebuild retrieval infrastructure from scratch.

Quivr supports multiple model providers, including OpenAI, Anthropic, Mistral, and local models through Ollama, and can work with common document types such as PDF, text, and Markdown. Its RAG workflows are configurable, allowing developers to change retrieval, reranking, conversation-history behavior, internet search, tools, and parsing, including integration with document-processing systems such as MegaParse.

The project is useful for developers building document assistants, knowledge bases, personal search tools, and other applications where RAG is a core component but not the product’s main differentiator. Quivr Core can be embedded into another application so teams can focus on their user experience and domain logic while reusing a maintained retrieval workflow.