About RAG Repo

RAG Repo is a curated directory of the best open datasets, knowledge bases, and data repositories for building Retrieval-Augmented Generation (RAG) systems: applications that answer questions by first retrieving relevant documents, then using a language model to write a grounded answer.

Why we built it

There is no shortage of open data. The hard part is knowing which sources are worth your time, how they are licensed, what shape the data comes in, and how much work it takes to get it into a vector database. That gap, between "the data exists somewhere" and "the data is ready to retrieve against", is where most RAG projects lose days.

We close that gap. Every source here has a plain-English description, a clear licence, its size and formats, and a note on how RAG-ready it is out of the box. You get enough to decide whether a source fits before you download a single byte.

How we choose sources

We favour accuracy over completeness. We would rather describe 80 sources well than list 200 badly. A source earns a place when it is genuinely useful for retrieval, openly accessible or clearly worth the access it requires, and properly documented so we can describe it honestly.

We always link to the primary source, never a mirror. We state the licence every time, and where a licence is unclear we say "check source" rather than guess. Our descriptions are neutral: we tell you what a source is and what it is good for, and let you decide. Our methodology pagesets out in full how we select, describe, and review every source.

Who maintains it

RAG Repo is built and maintained by Captivated Ltd. If you spot something out of date or have a source we should include, we would love to hear from you on our contribute page.