Quivr

Quivr is an open-source RAG tool that lets you ingest files like PDFs, Office docs, Markdown, CSV, audio and video via parsers such as MegaParse. You can create isolated 'Brains' for different knowledge domains, then simply @mention a Brain to get precise Q&A, summaries, translations or semantic search results within that scope.

It supports multiple LLMs, including GPT-4, Claude, Mistral, and local models via Ollama, giving you full control over data and privacy. It runs offline, and you can deploy with one click on Vercel/AWS or self-host entirely.

However, Quivr currently requires manual data uploads and lacks built-in real-time web search. Using commercial LLMs means managing API keys and paying usage fees. Non-technical users may find initial setup challenging. As an open-source project, long-term maintenance depends on the community.

It's ideal for researchers, developers, and privacy-focused teams comfortable with a DIY approach. It's less suited for those who need a plug-and-play, fully managed solution or real-time online information without manual uploads.