That link opens a working library of 91 example books: browse the shelf, open a book, ask the desk something. Nothing to install, nothing to sign up for, nothing saved to your device. Reload and it is gone.
When you want your own in it, open LibrAPP and photograph a shelf.
Point a camera at a bookcase. LibrAPP cuts the photograph up on your device and reads the spines into a catalog you can search, filter and browse. Bring in a spreadsheet or a barcode as well, and the same book arriving from three places becomes one entry rather than three.
Then ask it things. The LibrAPPrian's desk takes your own shelves as context: what to read next and why, which books by an author you have collected unevenly are missing, what threads run through what you own, a list for a long flight. The answers are about your books, not about books.
It also tells you things you did not ask. Which books you bought years ago and never opened, ranked by how long they have waited and by how much you evidently wanted them at the time. What is lent out and to whom. What the collection keeps coming back to.
Your books stay on your device. No account, no server, no sync. It works offline once loaded and installs as an app on Windows, Linux, Android and, by Add to Home Screen, on an iPhone. Three optional steps can send something, all of them listed below, and each shows you what it is sending first.
The interface is in English and Spanish, chosen on the opening page.
The rest of the manual lives in docs/, one page per thing.
| Adding your books | A photograph, a list you already keep, a Kindle library, a barcode, or typing one in |
| Using the catalog | Searching, the shelf view, favourites, corrections, lending |
| The LibrAPPrian's desk | What it works out on its own, what you can ask it, and what it sends |
| Where your library lives | Storage, backups, and moving to another device |
| Optional AI key | Which services, what they cost, and working without one |
| Browser support | What runs where, and what is missing |
| Development | Running it, testing it, and how the code is laid out |
| Import | shelf photographs, spreadsheets, CSV, barcodes |
| Merge | the same book from several sources becomes one entry, not duplicates |
| Browse | search, filter and group offline — no network, no waiting |
| Correct | edit or remove any entry; corrections outlast every rebuild |
| Ask | recommendations, synopses and reading lists, with your own shelves as the context |
| Complete | ask for the details the catalog is missing, and see them before they are kept |
| Mark | star the ones that matter and write your own notes, which the desk then reads |
| Track | who has the book you lent, and whose book you are still holding |
Two steps use AI: reading spines off a photograph, and asking the desk a question. Both work without a key, by preparing the request for you to paste into any AI session yourself, and both show what a request will cost before it is sent. Everything else, meaning importing, merging, searching and filtering, is plain code running locally. You choose which service to use and pay for it directly.
LibrAPP is available in English and Spanish.
Adding a language: add one file next to
web/src/i18n/en.js and listing it in
web/src/i18n/index.jsx. Any key you leave out falls
back to English, so a partial translation still works.
LibrAPP is built to be used by keyboard and by screen reader, not only by mouse and eye. It is checked against WCAG 2.2 rather than guessed at, and the colour arithmetic is measured by its own tests, so a later change to the palette has to fail there before it can ship.
Contrast is not a setting. The palette everyone gets is the raised one: quiet text reaches 7:1, which WCAG calls the enhanced level, and the hairlines between rows are drawn as strongly as the boundary of a control, since at low vision the structure of a page matters as much as the words. There used to be a switch offering a lower level; there is nothing to switch to now.
By keyboard, a skip link is the first thing Tab reaches, past the sidebar and into the catalog. The book panel is a real dialog: it takes focus when it opens, keeps it while open, closes on Escape, and hands focus back to the row you opened. The current page is marked as such, and the page title names the view you are on.
Everything that carries meaning carries it in words as well. Read state is written out, not only coloured. The genre chart is labelled and every slice is named and counted in its legend. Spines keep their titles as their accessible names, and word cloud entries say how many books use each word.
Nothing moves for anyone who has asked their system for less motion.
If something here does not work for you, that is a bug and worth reporting.
You do not have to install anything — the app runs in the browser. But installing gives it its own icon and window, and makes offline use reliable.
On your phone, the front page offers to install where the browser supports being asked, and writes out the manual route where it does not: on an iPhone, Share then Add to Home Screen; on Android, the browser menu then Install app. Both do the same thing.
Your catalog does not travel with it. Each device keeps its own, so moving one across is a file: export it in The stacks on one device, and bring that file in on the other. A backup works for this too, since a backup is the same file.
LibrAPP is a static site. Build it and serve the web/dist folder from
anywhere — a local server, a static host, GitHub Pages.
cd web && npm install && npm run buildRequirements: Node 20+ to build. Nothing to run it.
The opening page has a footer: About · Privacy · Licence · Source code · Report a problem. They all lead to one page, each landing on its own section, reachable from the sidebar once you are inside.
There is no contact form, because there is no server to receive one. Anything about the app goes to GitHub Issues, where it is public and does not get lost; anything else goes through the contact details on my own site.
There is no cookie banner either, because there are no cookies, no analytics and no trackers. The only things remembered about you are which language you chose and where your library is.
Three things can leave the device, all optional and none of them on by default:
| Step | What is sent | What is not |
|---|---|---|
| Reading a shelf photograph | The pieces of that photograph | Anything else about your catalog |
| Asking the desk a question | Your reading profile, printed in full first | The rest of the catalog |
| Scanning a barcode | The ISBN, to Open Library | The title, your notes, your shelf, you |
The barcode case is the smallest of the three by a distance. Barcodes are decoded on the device, whether they come from a picture or from the camera, and neither the picture nor the camera's view leaves or is recorded. What goes out is thirteen digits printed on the back of a book anybody can buy.
The camera is asked for only when you press Scan with the camera, and it is handed back the moment you stop, leave the page, or switch to another tab.
Version. The stacks shows which build you are running, when it was made, and a button that throws the cached copy away and fetches the current one. An installed app keeps a copy of itself so it can open offline, and that copy can occasionally be older than what is published. The button discards only the app — your library, sources and corrections are stored elsewhere and are not touched.
Written by Jesús J. Ballesteros working with Claude, an AI assistant. Every decision about what to build was human made; most of the code was typed by the model. A few early testers used it and said what did not work, and more than one thing here exists because of that.
None of this touches the contents of your catalog. No entry is invented, every book comes from a source you provided, and the AI features are optional and off until you add a key.
The Python scripts under tools/librapp/ predate the app and
still read and write the same folder layout. They are unmaintained: the app does
everything they did, and nothing in it depends on them.
cd web
npm install
npm run dev # development server
npm run build # production build into web/dist
npm test # the test suiteWhat the tests cover, how the code is laid out and why there are no browser tests: Development.
GNU Affero General Public License v3, or any later version.
Use it, change it, share it, for anything. Including in a business.
If you pass it on, pass the source on with it. That applies to a modified copy you host as well as one you hand over: this app is delivered to every visitor as JavaScript, so putting it on the web is handing it over. Publish your changes under the same licence and you have done what the licence asks.
A commercial exception is available. If you want to build LibrAPP into something closed, open an issue and ask. The copyright is held in one place, so it can be given.
LibrAPP was under PolyForm Noncommercial until v2.6.1. That licence barred commercial use outright, which also kept the project out of every free-software directory and off the list of things most people are allowed to contribute to. Versions up to v2.6.0 remain available under it.
Its dependencies keep their own terms: React, Zod, the Anthropic SDK and the barcode reader under MIT, the two typefaces under the SIL Open Font Licence. The app lists them on its About page.