Quivr

Quivr (specifically quivr-core) is the open-source Python library at the heart of the Quivr project, giving developers a batteries-included RAG pipeline they can drop into their own applications. It handles the usual chores of building a chat-over-your-docs system: ingesting PDFs, text, and Markdown files, chunking and embedding them, retrieving the right context, and routing the question to a chosen LLM. The pitch is that you can wire up a working assistant in roughly five lines of code, then progressively customize the pipeline with tools and internet search as your use case grows.

It is model-agnostic, with first-class support for OpenAI, Anthropic, Mistral, and Gemma, so teams can mix providers or swap them out without rebuilding the stack. It pairs naturally with Megaparse, the same team's document-parsing library, which matters if you are dealing with messy real-world PDFs. Pricing isn't a factor since the core library is free and installed via pip; you only pay for the LLM and any vector store you bring. This is squarely a developer tool, not a no-code SaaS, so expect to write Python and host the runtime yourself.

The trade-off versus hosted RAG platforms is the usual one: more flexibility and no per-seat fee, but you own the infra, observability, and evals. Teams that previously used the Quivr hosted product will recognize the philosophy, but core is the library layer rather than a full UI.