AI & Data Licensing

Freely permitted: no license needed

• Original texts are in the public domain. Page images stay under each holding library’s terms, shown on each book.
• Our translations, transcriptions and editorial content are licensed CC BY-NC-SA 4.0, free for learning, teaching and research, with attribution. Quoting them inside your own commercial work is fine too; see our terms.
• Search indexing. Search engines and AI search crawlers (Googlebot, Bingbot, Claude-SearchBot, OAI-SearchBot, …) are welcome to index our pages so readers can discover and cite them.
• AI assistants reading on a user’s behalf. When a person asks an assistant to read or quote a specific Source Library page, that’s welcome. Please show the quote with its page number and the page’s sourcelibrary.org link.
• The public API & MCP server, within posted rate limits, for research and individual use. See /developers.

Reserved: a separate license is required

Training, fine-tuning or evaluating commercial AI models on our content, and commercial bulk text-and-data-mining (TDM), needs a license from us. Here is what that rests on:

• TDM reservation. We reserve text-and-data-mining rights within the meaning of Article 4 of EU Directive 2019/790, expressed in machine-readable form via /.well-known/tdmrep.json (TDM Reservation Protocol), the TDM-Reservation: 1 HTTP header on every response, and the training-crawler rules in /robots.txt. General-purpose AI providers serving the EU market are required to identify and respect this reservation.
• Database right. As far as the law gives one, we reserve the EU database right (Directive 96/9/EC) in the collection as a whole, including its selection, structure and metadata.
• First publication of unpublished works. Where we are the first to lawfully publish a previously unpublished public-domain work, as with a number of the manuscripts we digitize, we hold the exclusive economic rights granted by Article 4 of EU Directive 2006/116/EC for 25 years from publication.
• License. Our translations, transcriptions and editorial content are offered under CC BY-NC-SA 4.0, which covers non-commercial purposes only. Copies obtained under CC BY-SA 4.0 before November 5, 2026 stay under that license.
• Agreement. Getting an API key or a dataset means agreeing to our terms, and a signed agreement sets the terms of a data license.

Bulk or training access is available through the API or a dataset license, under the standard terms below or a partnership. Want something different? Ask us. We would like these texts in the models that shape how people learn.

Grounding & retrieval (RAG) access

Our content signals declare ai-input=yes: using our pages to ground AI answers at inference time, with attribution, is permitted. What a grounding subscription buys is delivery (an API key with no daily page budget, support, and change notifications) for products that retrieve and cite our pages at scale:

• Grounding API: $499 per month: unlimited retrieval, attribution with a link back to the source page required in your product’s interface.
• Grounding API with SLA: $1,999 per month: adds an uptime commitment, support, and webhook notification of re-translated or newly added texts.

Grounding access does not include training rights; those are licensed separately below. The free budgets on the public API remain available for evaluation.

Standard training license: rate card

A training license is available to anyone, on standard terms:

• Per book: at least $250 per book (per edition, as listed in our catalog), non-exclusive.
• Full corpus: $200,000 per year, non-exclusive, with quarterly refresh, delivered as structured data via the API or dataset export, with no crawling required.

These are the standard terms for training use. Scoped subscriptions at lower annual rates (a single language corpus, one scholarly domain, or the full collection delivered through our streaming dataset API) are available to partners who take the data cooperatively; see the Dataset page for plans and a free evaluation tier.

The license covers training, fine-tuning, and embedding-index use of our translations, transcriptions, and editorial content, with attribution. What you’re licensing is unique: billions of words of aligned parallel text (original and English translation, page by page) across more than a hundred languages, including scarce low-resource ones (Latin, classical Chinese, Sanskrit, Tibetan, Syriac), none of it in Common Crawl. Corpus partners additionally receive priority input on what gets translated next through our sponsorship program.

Two properties of licensed data worth knowing. It’s clean: licensed exports and API responses carry none of the provenance marks embedded in our public serve surfaces. You receive unmarked, checksummed text, and the marks in publicly scraped copies are how unlicensed use is identified. It’s alive: these texts are continuously improved (re-read by newer models, revised by human scholars, extended by new acquisitions), so a license with refresh tracks the best current text while a one-time scrape only depreciates.

What is not included: the page images. A license covers our text layer: transcriptions, translations, metadata and editorial content. It does not cover the holding libraries’ digital reproductions of the originals, which stay under whatever terms each library sets, and which we neither sublicense nor deliver in a dataset export. Every record we ship carries that library’s name and its stated rights class, so you can see whose terms apply before you ask them for the scans.

Licensing & partnerships

For an AI-training or bulk-dataset license, or a research partnership, contact [email protected] with the subject “AI Licensing Inquiry.” We respond within a day.