Quivr provides an opinionated, fast, and efficient Retrieval-Augmented Generation (RAG) framework designed to simplify the integration of generative AI into existing applications. It targets developers who want to focus on their product rather than the complexities of RAG implementation, offering broad compatibility with various LLMs, vector stores, and file types.

How It Works

Quivr abstracts the RAG pipeline, allowing users to ingest various file types (PDF, TXT, Markdown) and query them using natural language. It supports multiple LLM providers (OpenAI, Anthropic, Mistral, Gemma) and local models via Ollama, as well as vector stores like PGVector and Faiss. Users can customize the RAG workflow, including adding internet search capabilities and tools, by defining configurations in YAML files.

Quick Start & Requirements

• Install: pip install quivr-core
• Prerequisites: Python 3.10 or newer. API keys for LLM providers (e.g., OPENAI_API_KEY) need to be set as environment variables.
• Documentation: https://docs.quivr.app/

Highlighted Details

• Supports integration with Megaparse for file ingestion.
• Customizable RAG workflows via YAML configuration.
• Allows for adding internet search and custom tools.
• Offers broad LLM and vector store compatibility.

Maintenance & Community

The project lists contributors and has a project board for tracking development. Links to community channels or social media are not explicitly provided in the README.

Licensing & Compatibility

Licensed under Apache 2.0, which is permissive and generally compatible with commercial and closed-source applications.

Limitations & Caveats

The README indicates ongoing development and future feature additions, suggesting the project may still be evolving. Specific details on performance benchmarks or advanced customization limits are not detailed.