Quivr Core
Python RAG framework for document-based AI apps, with local models through Ollama, cloud APIs, and customizable retrieval workflows.
Quivr Core is a Python framework for developers adding document-based AI answers to their own applications. It combines file ingestion with retrieval-augmented generation (RAG), so a model can answer questions using material from your documents. It supports local models through Ollama as well as cloud APIs from OpenAI, Anthropic, and Mistral.
The framework provides a starting workflow while leaving retrieval behavior open to customization. You can adapt how it handles conversation history, rewrites questions, retrieves relevant passages, and generates answers. It also supports reranking through Cohere and streaming answers to users.
Quivr Core accepts PDF, TXT, and Markdown files, with custom parsers for other formats. Its Megaparse integration provides another route for document ingestion. PGVector and Faiss are supported vector stores, giving developers a choice of where to keep the searchable document index.
Model choice affects where processing happens: Ollama runs models locally, while the supported cloud APIs send model requests to external providers. Internet search and tool use can extend the document workflow when an application needs information or actions beyond its files.
The Apache 2.0 licensed core library can be embedded in an existing Python application. Developers can change retrieval strategies independently of the application's user interface.
