Project Gutenberg

Project Gutenberg began in 1971, which makes it the oldest digital library on the internet. Volunteers digitise, proofread, and archive books whose copyright has expired, giving you clean full-text works you can build on without licensing worries. The catalogue runs to more than 70,000 titles, roughly 55,000 of them in English, with thousands more across dozens of other languages, from classic novels and poetry to reference works, plays, and historical texts.

Individual books are available as plain text, EPUB, HTML, and Kindle files. For bulk work, Project Gutenberg asks you not to hammer the main website with a crawler and points you instead to its mirror sites, which let you download the whole collection politely. The plain-text editions are the easiest to chunk (split into passages) and embed for retrieval.

A practical tip: older files wrap the actual book in a standard licence header and footer, plus the occasional transcriber note. Strip these before you index, or your retrieval will surface boilerplate instead of prose. If you would rather skip that cleanup, the Standardized Project Gutenberg Corpus republishes the collection with consistent formatting, metadata, and token counts.

For RAG, Gutenberg is the go-to source of long-form, public domain literature: well suited to question answering over classic texts, literary and stylistic analysis, character search, or clean training text where licensing must be beyond doubt. It fits any project where copyright cleanliness matters more than currency.

The main caveat is jurisdiction. Copyright is assessed per work and against United States law, so a title that is public domain in the US may still be in copyright where you are. Text quality also varies: most books are carefully proofread, but scanning artefacts and inconsistent formatting turn up in older entries. It sits alongside Open Library and the Internet Archive when you need a wider catalogue.