OUR STORY

A small library
with a curious mind.

The Free Library helps you find stories worth investigating.

Our shelves combine 350 English-language crime and espionage classics with a much larger catalogue of modern novels, story collections and selected plays. Explore Christie, Asimov, Wilde, Greene, le Carré, Grisham, Harry Potter and Slough House, alongside fiction from the bestseller archive. Gutenberg entries link to reading formats; modern entries link to book information.

Start with a few books you’ve enjoyed. Your reactions help us follow connections to unfamiliar stories and other authors. You can also browse the shelves or keep a reading list, without an account.

How recommendations begin

Choose “don’t like”, “like”, or “like a lot” for books you know. Unrated books remain unknown. You can change or remove a rating at any time; your choices and the calculations stay in your browser.

We compare the ideas in each available description. The original 350 classics also have twelve sampled passages spread through each book. A separate model looks for connections in the writing style of those passages. These compact numerical representations are called embeddings. They work alongside theme descriptors and simple prose measurements. Several starting taste profiles are blended from your ratings, and a regularised personal model adjusts as you add reactions.

Different sides of your taste can coexist. The results connect each suggestion to a book you liked and include other authors rather than filling the list with a single name. Matching does not use an author’s popularity, social following or subscription exchanges.

A first model, with room to learn

The starting profiles are editorial hypotheses. The expanded books use attributed encyclopedia descriptions or our original descriptions and theme descriptors. Entries awaiting a checked story description can be browsed, but do not enter recommendation matching. For The Picture of Dorian Gray and Crime and Punishment, we also analyse three 250-word passages from the opening. The latter uses Constance Garnett’s English translation, whose wording shapes the style comparison. Modern books, including Eco, Potter and Slough House, use descriptions only. Twelve sampled passages do not represent every part of a novel, and a similar style or subject does not establish that you will enjoy it.

The pretrained representations use BGE for content and StyleDistance for style. Passage analysis runs locally during catalogue preparation. The website receives numerical profiles, synopses and source links.

The research behind the longer-term approach includes MeLU: Meta-Learned User Preference Estimator for Cold-Start Recommendation, which learns a starting model from previous readers and adapts it using a few ratings. Our launch model has no shared reader-training data and is not a trained MeLU system. Its predictions still need evaluation against readers’ reactions to books withheld from their profiles.

Independent writers

We want to help contemporary crime, spy and speculative mystery writers reach readers who enjoy their writing. Future entries can link to the author’s own reading or purchase page.