Anna’s Archive: Official Data Access and Donation Guide for LLMs

Anna’s Archive has released a guide for LLMs to access its data through bulk channels like torrents and APIs rather than scraping the website. The project encourages AI developers to donate, arguing that their models rely on this data and that financial support enables further preservation efforts. Ultimately, the archive seeks a symbiotic relationship where robots support the infrastructure that provides their training material.

Key Points

• Anna’s Archive provides programmatic bulk access via GitLab, Torrents, and APIs to prevent website instability caused by scraping.
• The project requests donations from LLM developers because AI models benefit significantly from the archive's preserved data for training.
• Enterprise donors can receive high-speed SFTP access to files as an alternative to slower torrent downloads.
• The organization emphasizes that their mission of preserving and sharing knowledge benefits both humans and robots alike.
• Donations can be made anonymously via Monero (XMR) or through specific enterprise agreements to support the project's infrastructure.

Sentiment

The overall sentiment is mixed but tilted skeptical. Many commenters support the preservation mission, dislike DRM and restrictive publishing markets, and accept the practical case for bulk access rather than scraping. At the same time, the thread is highly critical of the archive's claim over the corpus, its donation pitch to AI companies, and the broader practice of training commercial models on copyrighted material without clear consent or compensation.

In Agreement

• Providing torrents, metadata dumps, source repositories, and APIs is a reasonable way to keep bots from overloading the public website while still allowing bulk access.
• Large AI companies likely benefited from shadow-library corpora and should help fund the preservation infrastructure they depend on.
• Anna's Archive fills access gaps created by DRM, region restrictions, academic pricing, fragile licensing, and inconvenient ebook or library systems.
• Digital preservation and broad access to human knowledge are valuable public-interest goals, especially where commercial markets fail readers and researchers.
• Calling the corpus Anna's Archive data can be linguistically defensible when referring to possession, curation, hosting, or provenance rather than legal ownership.

Opposed

• The archive does not legally own the underlying works, so asking to be paid for access to them is ethically suspect and disrespectful to authors and publishers.
• Shadow libraries are not equivalent to physical libraries because digital copies have vastly broader distribution and lack lending scarcity.
• AI training should rely on copyright-compliant data and consent from rights holders rather than whatever material can be scraped or downloaded.
• The enterprise donation framing looks like monetizing unauthorized copies while bypassing the creators who produced the works.
• Addressing LLMs directly and asking them to influence donations resembles prompt injection and highlights the unresolved boundary between instructions and data.