Welcome to RAG Repo
RAG Repo is a free, curated directory of open datasets and knowledge bases for building RAG systems, plus plain-English guides to help you use them.
Welcome to RAG Repo. We built this to answer a question we kept running into ourselves: when you want to build a Retrieval-Augmented Generation (RAG) system, an AI that looks things up before it answers, where do you actually get the data to feed it? There is a lot of open data in the world, but finding the right pieces, and working out whether they are ready to use, can eat up days. We wanted a single, trustworthy place to start. So we made one.
What RAG Repo is
At its heart, RAG Repo is a curated directory of data sources: open datasets, knowledge bases, and public repositories that are genuinely useful for building RAG systems. Each one has its own profile page telling you what it contains, who maintains it, what licence it comes under, how big it is, what formats it ships in, and how ready it is to drop into a retrieval pipeline. You can browse everything in the directory, filter by category or access type, and search for what you need.
We are opinionated in a useful way. Rather than listing every dataset that exists, we pick the ones worth your time and tell you plainly what each is good for. Think of us as a knowledgeable friend pointing you towards the good stuff, not a search engine handing you 10,000 results and wishing you luck.
Why “data exists” and “data is RAG-ready” are not the same thing
Here is the gap that started all of this. There is an enormous amount of open data out there. Common Crawl alone publishes hundreds of terabytes of web pages every month. Wikipedia, arXiv, government portals, legal archives, code repositories: the raw material is abundant and, in many cases, free.
But raw data and RAG-ready data are two different things. To use a source in a RAG system, you usually need to clean it, split it into passages small enough to retrieve (a step called chunking), and often turn those passages into embeddings, which are numerical representations that let a system find text by meaning rather than by keyword. Some sources arrive already cleaned and chunked, and a few even come pre-embedded. Most do not. A 45 TB web archive is a treasure, but it is not something you plug straight into a vector database on a Tuesday afternoon.
So the questions that actually matter are practical ones. Is this licensed in a way that lets me use it? Is it structured text or a pile of raw HTML? Will it fit my budget and my hardware? Is there an API, or do I need to download the whole thing? These are exactly the things we try to answer on every profile page, so you can judge the effort before you commit to it.
The .org ethos
We are rag-repo.org, and that is deliberate. RAG Repo is a community resource first, not a sales funnel. We do not take payment to feature a source, we do not rank listings by who pays us, and we will never bury the good free option under a sponsored one. Our editorial choices, which sources we include, how we describe them, and which we highlight, are ours alone and are never influenced by advertising.
To keep the lights on, you will see a single, quiet advertisement on some pages. One per page, clearly separated from the content, never dressed up to look like a recommendation. That is the whole business model for now: enough to cover hosting and the domain, and nothing that gets in your way. You can read more about who we are and how we work on our about page.
What is here today
Three things, to start.
The directory. The core of the site. Every source we have catalogued, browsable and searchable, with a detailed profile for each. This is the part you will probably use most.
Learn. Plain-English guides in the Learn section that explain the ideas behind RAG without assuming you already know them. What RAG actually is, how chunking works, what embeddings are, how to choose a vector database, and how to read a data licence so you do not get caught out later. We define the jargon as we go, because understanding a field should not depend on already being inside it.
The blog. This post is the first entry. We will use this space to share new additions to the directory, notes on interesting datasets, and thoughts on building with RAG as the field moves.
That is where we are on day one. There is more to come: community submissions, collections that bundle related sources together, and eventually a marketplace for curated, RAG-ready data. But the foundation is here now, and it is free and useful today.
Have a look around, start with the directory, and thanks for stopping by. We are glad you are here.