Digitised book collections, literary corpora, and audiobook archives, primarily public domain works.
6sources
Books and literature datasets are large collections of long-form text: novels, non-fiction, poetry, and reference works, most of them old enough to be in the public domain (free of copyright, so anyone can reuse them). They matter for a RAG (Retrieval-Augmented Generation) system because books give you something web pages rarely do: sustained, carefully edited prose on a single subject. That makes them excellent for chunking, the process of splitting text into passages, because the writing already flows in coherent, self-contained sections.
Choosing within this category comes down to how much cleaning has been done. Raw archives hand you the full text of hundreds of thousands of titles, but often wrapped in headers, licence notices, and inconsistent formatting you will need to strip out. Cleaned corpora have already removed that boilerplate and standardised the text, saving you the tedious work at the cost of some flexibility. Catalogue and metadata sources, meanwhile, are less about full text and more about knowing what exists.
Two things to keep in mind. First, public domain status varies by country, so a title that is free to use in one place may still be in copyright in another. Second, older books carry older facts and dated language, which is fine for literary or historical work but a poor fit if you need current information. Watch too for scanning errors in digitised text, which can quietly corrupt passages.
The sources below cover full-text archives, cleaned corpora, catalogues, and spoken-word recordings of public domain works.
Word and phrase frequency data drawn from Google's digitised book collection, spanning centuries of published text. It counts how often words and short phrases appear by year, which is useful for historical language analysis and time-aware RAG features.
A 242 billion token dataset of roughly 983,000 public-domain volumes digitised from Harvard Library's collections, spanning more than 250 languages, with both raw and post-processed OCR text plus rich bibliographic metadata.
A volunteer project providing free public domain audiobooks. Volunteers record themselves reading works whose copyright has expired, which makes it a handy source of speech data for training speech-to-text models and building multimodal RAG.
An Internet Archive project building a web page for every book ever published. It holds more than 20 million catalogue records and lends many titles digitally, making it a rich source of book metadata for RAG systems.
A volunteer effort to digitise and archive public domain books, with more than 70,000 free ebooks. Mostly English, but it covers many languages. One of the oldest digital library projects, so it is a clean, permissively licensed source of full-text literature for RAG.
A research-ready version of the Project Gutenberg catalogue with consistent formatting, tidy metadata, and token counts for every book. It saves you the cleanup work, so you get uniform full-text literature ready to chunk and embed for RAG.