AI cataloging Cataloging
Photograph the title page and copyright page. The fields fill in, each with its source — ready for you to review and save to Koha as a MARC21 record.
AI Cataloging for Koha libraries
Catalog a book by photographing it.
Works with the Koha in your building, even when the server has no internet. Photograph the title page and copyright page: CatalogLense reads them, checks them against the Library of Congress, and saves the record and the copy (barcode, call number, shelf) to your Koha in one step, every field showing where it came from.
Android and iOS Works with Koha 19.05+ Runs on a local-network Koha
‹
Validate Bibliographic DataLAN · koha.library.net
1 AI value not found on the pages
245 $aTitleLoC
Noli me tángere
100 $aAuthorLoC
Rizal, José, 1861-1896.
020 $aISBNscan
9789712731938
650 $aSubjectLCSH
Philippines -- History -- 1812-1898 -- Fiction.
490 $aSeriesAI
Philippine Classics Library
Not found on the photographed pages — check this value.
Save to Koha
The backlog
Thirty minutes of typing. Now it's a phone camera.
Most school, barangay and parish libraries don't have a cataloger on staff, so every new book waits for someone to type its record by hand. Work out what your own pile costs.
An estimate: photographing a few pages, then checking the fields marked for review. Measure your own during a pilot.
100 books typed by hand is 8.3 working days.
6.9working days back
12books a day, typed
72books a day, reviewed
A working day here is the hours you chose above. The CatalogLense figure is your estimate, not a measurement.
Words on this page
- Original cataloging n.
- Making a catalog record from scratch, by reading the details off the book itself. What CatalogLense speeds up.
- Copy cataloging Z39.50
- Importing a record another library already made, over the Z39.50 search protocol. Only works if someone cataloged the book first.
- MARC record MARC21
- The standard format for a catalog record. Each piece of information sits in a numbered field: 245 is the title, 100 the author, 650 a subject.
Z39.50 copy cataloging
Local publications, older stock and donated textbooks come back empty — and it needs your Koha to reach the internet.
ISBN lookup tools
Books printed before ISBNs, locally published titles and school editions often don't have one.
Typing it into Koha
Works — at a desktop, away from the shelf, a few books an hour, while the cart refills.
CatalogLense reads the book itself, so it doesn't depend on someone else having cataloged it first.
How it works
From the shelf to Koha in five steps.
Tap a step to see it on the phone. Only one step needs you — the review — and that's on purpose.
Copyright page
The back of the title page — ISBN, publisher, place and year.
• The ISBN is read from here, so make sure the ISBN line is sharp.
‹
Reading pagesEach page read for the fields it's the source for
Title page✓ read
245 title · 100 author · 250 editionCopyright page✓ read
264 publisher, place, date · 020 ISBNTranscript checkrunning…
"Anvil Publishing, Inc." · "Pasig City" · "2011"Accession stamppublisher
Ignored — a library stamp isn't the publisher.‹
Cross-checkingSearched in parallel
Library of CongressLoC
Record found — overrides 245, 100, 264LCSH authorityLCSH
Philippines historyPhilippines -- History -- 1812-1898 -- Fiction.
Open LibraryOL
Added 2 subjectsGoogle BooksGB
Filled 1 field that was still empty‹
Validate Bibliographic DataReview before saving to Koha
1 AI value not found on the pages
245 $aTitleLoC
Noli me tángere
100 $aAuthorLoC
Rizal, José, 1861-1896.
020 $aISBNscan
9789712731938
490 $aSeriesAI
Philippine Classics Library
Not found on the photographed pages — check this value.
Continue to copy details
Saved to Koha
Under your own Koha account
Recordbiblionumber 18342
Copy barcode0001245
Call number899.2113 R52n 2011
Shelving locationGeneral Collection
Catalog the next book
Capture the pages that count.
The camera asks for each page by name, with a framing guide shaped for it and tips for a sharp shot: the title page first, then the copyright page, then optional extras like the CIP block, back cover, contents and spine. Each is captured at its own resolution, so 6-point copyright type gets the pixels it needs.
- Title page AACR2 · RDA
- The page the cataloging rules treat as the book's own statement of its title and author. The cover isn't — it's often shortened or restyled.
- CIP block Cataloging in Publication
- A small block of catalog data some publishers print on the copyright page. A head start, when the book has one.
The pages are read.
The photos go to the AI, told which page it's looking at and which fields that page is the source for. That's why a library's accession stamp doesn't get read as the publisher.
- AI credit n.
- One book read by the AI. Credits come with your plan, so there's no Google or Anthropic account to set up. A read that fails costs nothing.
- Transcript n.
- A word-for-word copy of the printed text on each photo. Every value the AI reads is later looked for in it.
Checked against real catalogs.
The Library of Congress, Open Library and Google Books are searched at the same time. A Library of Congress record wins over what the photos said, and subject headings are matched to the authorized form.
- Library of Congress record
- The Library of Congress's own catalog entry for the book, when one exists. The most trusted outside source there is.
- LCSH LC Subject Headings
- The controlled list of subject terms most libraries use, so a topic is always filed under the same words.
The step that needs a cataloger
You review every field.
About 34 MARC fields, each marked with where it came from. Anything the AI read that isn't printed on the pages is flagged. Fix what's wrong, get warned if you already hold the title, then add the copy: barcode, call number, shelving location.
- Field and subfield 245 $a
- A field is a numbered slot in the record — 245 is the title. A subfield is a labeled part of it: $a the title proper, $b the subtitle.
- ISBD punctuation
- The standard punctuation between parts of a record, like " : " before a subtitle. Applied for you.
Saved straight into Koha.
The record and the copy are created through Koha's REST API, with your own Koha account and permissions. If the server can't be reached, the save waits on the phone and goes through when it's back.
- Record and copy biblio · item
- The record describes the book. Each copy is one physical book, with its own barcode and shelf location.
- Koha REST API
- The door Koha provides for apps. It has to be switched on by your Koha administrator — see System requirements.
Why the records can be trusted
An AI that's confidently wrong is worse than a blank form.
A catalog is a permanent record. Four safeguards make sure a guess never looks like a fact — try each one below.
Every field says where it came from.
Each value carries a badge naming its source, and the sources are ranked. A subject the AI guessed from cover art never looks as solid as a title from the Library of Congress — so your review time goes to the fields that need it.
- Provenance n.
- A record of where something came from. Here: which source supplied each field.
The trust ladder — tap a source
Every AI value is looked for on the page.
Each photo is transcribed word for word, and every value the AI read is searched for in that transcript. A value that isn't printed anywhere is flagged for you. Nothing is deleted automatically — you decide.
- Grounding n.
- Tying each answer back to the text it came from. An answer with nothing to tie it to is suspect.
Transcript · copyright page
Copyright © 2011 by Anvil Publishing, Inc. All rights reserved. Published by Anvil Publishing, Inc. Pasig City, Philippines First printing, 2011 ISBN 978-971-27-3193-8
- 264 $b Anvil Publishing, Inc.
- 264 $a Pasig City
- 264 $c 2011
- 020 $a 9789712731938
- 490 $a Philippine Classics Library
Rules catch what rules can prove.
Some mistakes don't need AI to find. ISBN check digits, impossible years, language names where a MARC code belongs, LCCN and Dewey formats — all checked on the phone, without AI and without a network.
- Check digit ISBN
- The last digit of an ISBN, calculated from the others. One wrong or swapped digit, and the math no longer adds up.
"Filipino"→fillanguage code2099→ flaggedyear in the future
Editing a record can't destroy it.
When you edit a record, only the fields you changed are rewritten. Control fields, Koha's own fields, your local notes and every subfield you didn't touch pass through byte for byte — the one piece of logic backed by its own automated tests.
- Byte-identical adj.
- Exactly the same, down to the last character. Not "equivalent" — the same.
A record in Koha
- 001 18342
- 005 20260921093012.0
- 020
$a9789712731938 - 100 1_
$aRizal, José,$d1861-1896. - 245 10
$aNoli me tángere /$cJosé Rizal. - 250
$a1st ed. - 264 _1
$aPasig City :$bAnvil Publishing,$c2011. - 590
$aDonated by the Class of 1998. - 650 _0
$aPhilippines$xHistory$y1812-1898$vFiction. - 942
$cBK - 999
$c18342$d18342
Nothing edited yet.
Nothing reaches Koha without a person.
CatalogLense assumes a cataloger reviews every record. It doesn't catalog unattended, and it never will.
What's in the app
Useful for the rest of the day, too.
AI cataloging is the headline, but a cataloger's phone earns its keep at the shelf, the desk and the circulation counter. Each tile shows the feature at work.
Shelf reading Inventory
Scan a shelving location book by book: on shelf, misplaced, not in catalog, lost or withdrawn — and at the end, what never turned up.
Add another copy
Find the record, attach a new copy. No recataloging.
Manual cataloging
Rather type it? A form with your framework's fields, timed against AI on your own phone.
Catalog search
Keyword, title, author, ISBN or accession number.
Circulation desk New
Patron lookup; check out, renew, check in, holds and payments. Photos with the CatalogLense plugin.
Barcode scanning
Auto-detects the format, or lock it to your labels'. Handheld scanners work too.
Reports New
How many items your Koha holds under each collection code and shelving location, and your cataloging pace on this phone. For today, this month, this year or all time.
Offline queue
Koha unreachable? Saves wait on the phone and upload when it's back. No signal for the AI? Photos wait too.
Call numbers
Built from Dewey, author cutter and year, ready for the spine label.
Record maintenance
See full MARC and copies; edit or delete either.
Cataloging defaults
Set once, applied to every book you add.
Several libraries
Save each library's Koha and switch between them. Each keeps its own login, defaults and history.
Correction log
What the AI got wrong and what you fixed — exportable as JSON.
Permission check
See what your Koha account can do right after sign-in.
Made for offline libraries
Built for the Koha server in your building.
Most school and small public library Koha installs are a machine down the hall, reached by a bare IP address. No domain name, no certificate, often no internet. Tools built for hosted Koha fail at the first screen. This one starts there.
A bare IP is a supported address.
Plain http:// is accepted for addresses the internet can't reach — private ranges, .local names — and refused for public hosts, where your password really would be exposed.
Sign-in adapts to your server.
The app checks whether your Koha offers password or API-key sign-in, and prefers API keys: they expire on their own and can be revoked per phone.
A failed connection says what's wrong.
A missing RESTBasicAuth setting, an untrusted certificate and a blocked local-network permission look the same from inside most apps. CatalogLense names the likely cause.
Koha server address
http://192.168.1.50:8080Local network — acceptedhttp://koha.localLocal name — acceptedhttps://library.example.edu.phEncrypted — acceptedhttp://library.example.edu.phPublic and unencrypted — refused, to protect staff passwordsThe honest boundary: reading the photos needs internet on the phone, because the AI is a cloud service. Photographing doesn't: books shot with no signal wait on the phone and are read once it's back. Signing in to your CatalogLense account the first time needs internet too. Everything else — review, saving, search, shelf reading, circulation — is between the phone and your own Koha. Your Koha server never needs to be online.
Compared with cloud tools
On your network, not someone else's cloud.
Cloud cataloging services work from a desk and need to reach your Koha over the internet. CatalogLense works at the shelf, on the network your Koha is already on.
Typical cloud tools
CatalogLense
CatalogLense
A Koha on your local network
Their cloud has to reach Koha over the internet
The phone talks to Koha directly; the server can stay offline
The copy
The record only; barcodes and call numbers are added in Koha
Record and copy together: barcode, call number, shelf
Where you work
At a desk, in a browser
At the shelf, with the book in your hand
Price
Usually in US dollars only, per credit
Free while in beta, then priced for your country
The rest of the day
Cataloging only
Circulation, shelf reading and reports too
System requirements
Check in ten seconds.
What the phones, the network and your Koha server need. CatalogLense works inside the permissions your library already grants — it never asks for administrator rights and never changes a system preference.
Your Koha's REST API must be enabled
CatalogLense can't work with your library until the REST API is turned on in Koha's Administration panel. In the Koha staff interface, go to Administration → Global system preferences → Web services and enable RESTBasicAuth, RESTOAuth2ClientCredentials, or both. If you don't administer Koha yourself, ask your Koha IT administrator — it takes a minute, and the app can't do it for you.
Phones and tablets Staff devices
- Android
- Android 7.0 or newer
- iPhone and iPad
- iOS / iPadOS 16.4 or newer
- Camera
- A rear camera with autofocus, for small print on copyright pages.
- Barcode scanner Optional
- The phone camera reads barcodes. A handheld scanner that types like a keyboard works too.
Network Where the phone connects
- To your Koha
- The phone must reach the Koha server — usually the library's staff Wi-Fi. A bare local IP like
192.168.1.50is fine. - Internet On the phone
- Once, to sign in to your CatalogLense account; after that, only while the AI reads photos. Your Koha server never needs to be online.
- Offline
- Saves made while Koha is unreachable wait on the phone and upload later.
Koha server Set by your administrator
- Version
- Koha 19.05 or newer
- REST API
RESTBasicAuthorRESTOAuth2ClientCredentialsenabled — see above.- CatalogLense plugin Optional · free
- Adds two things Koha's REST API lacks: uploading cover images, and showing patron photos (with the
patronimagespreference on). Needs Koha 22.11 or newer; installed under Administration → Manage plugins. Everything else works without it, and you can skip covers if your Koha already gets them from Amazon or a Coce server. Download the plugin (v1.1.2, .kpz, 52 KB)
Accounts Per library
- Koha staff account
catalogueto search and view;editcatalogueto create records and copies;circulateto check out, check in and renew;borrowers(view) to look up patrons. With the plugin,tools(upload local cover images) for covers andborrowers(list) for patron photos. Each is only needed for its own feature, and the app shows what the account can do after sign-in.- CatalogLense account
- One per person, by email — sign-in is a code we email you, no password. The library's owner holds the plan; on Pro they invite up to 4 staff.
- AI Nothing to set up
- Credits come with every plan. On Free you can use your own Google Gemini or Anthropic Claude key instead, kept in the phone's secure keychain.
What it doesn't do
Plainly, so you can decide quickly.
- It doesn't catalog unattended. Every record passes a review screen.
- Reading photos needs internet. You can photograph and save offline; the AI reads the photos once the phone is back online.
- It doesn't replace Koha. It's a phone app for the Koha you already run.
- It reads books. The field set covers monographs, not serials or non-print.
- Photo quality matters. Glare, a curved page or a steep angle costs fields. The app guides each shot and flags photos too small to read.
- Book photos leave the phone — through CatalogLense to the AI (Google or Anthropic), or straight to it with your own key. Nothing from your Koha does.
What's coming
Being worked on next. No dates we can't keep.
- Offline shelf reading — a whole shelf session with no network.
- Bigger shelves — sessions beyond 500 items in one pass.
- A web app — CatalogLense in a desktop browser, for cataloging at the desk.
Plans
What each plan includes.
CatalogLense is in beta: Starter and Pro are free while we test. Every plan has the whole cataloging workflow. Plans differ in AI credits, who can sign in, and two extras. One credit is one book read by the AI.
Free
To try itAlways free
15 books to try it
- AI credits
- 15, once — valid 12 months
- Or your own AI key
- Google Gemini or Anthropic Claude, no credit limit
- Who signs in
- 1 person
- Accuracy reports
- Always on
- Cover upload, shelf reading
- —
Starter
One catalogerFree while in beta
100 books a month
- AI credits
- 100 every month
- Who signs in
- 1 person
- Accuracy reports
- On by default, can turn off
- Cover upload, shelf reading
- Yes
- Top-up credits
- Yes
Pro
For a teamFree while in beta
500 books a month
- AI credits
- 500 every month
- Who signs in
- The owner + 4 staff
- Accuracy reports
- On by default, can turn off
- Cover upload, shelf reading
- Yes
- Top-up credits
- Yes
Enterprise
Library systemsCustom
Sized for your libraries
- AI credits
- A volume that fits your collection, backlog projects included
- Who signs in
- Staff across your branches
- Billing
- By invoice, for a school, city or consortium
- Setup help
- We help your IT switch on Koha's REST API and install the plugin
- Support
- Priority email, and training for your catalogers
Beyond the app
- Koha in the cloud — we set up and run Koha for you, ready for CatalogLense.
- Systems around Koha — we propose other systems that work with your Koha, fitted to how your library runs.
- More Koha plugins — plugins that connect your Koha to those systems.
Every plan: manual cataloging, AI vs manual comparison, barcode scanning, catalog search, record editing, adding copies, patron search, checkout, reports, saved servers for several libraries, the offline queue, photos waiting for AI, the permission check, and the in-app guide.
Top-ups — +100 or +500 credits when a month runs short, valid 12 months.
Credits reset monthly and don't carry over. A read that fails costs nothing.
After the beta, plans are bought in the app through the App Store or Google Play, in your own currency. Prices will be announced at launch.
If you stop subscribing, your records stay in your Koha. They were never anywhere else.
Help us test it. Testers use Starter free during the beta and tell us what works and what doesn't. Best if your library runs Koha.
Join the betaQuestions
Questions librarians ask.
What is CatalogLense?
A phone app for Android and iOS that catalogs books for Koha. You photograph the title page and copyright page; it builds a MARC21 record, checks it against the Library of Congress, Open Library and Google Books, and saves the record and the item (barcode, call number, shelving location) to your Koha once you've reviewed it.
Which Koha versions does it work with?
Koha 19.05 or newer, with the REST API switched on (RESTBasicAuth or RESTOAuth2ClientCredentials). The optional CatalogLense plugin, for cover images and patron photos, needs Koha 22.11 or newer.
Does my Koha server need to be on the internet?
No. The phone talks to Koha directly over your library's network, and a bare local IP address like 192.168.1.50 works. The phone itself needs internet only while the AI reads photos, and to sign in the first time.
Can it catalog books that have no ISBN?
Yes. It reads the title page and copyright page themselves, so books printed before ISBNs, local publications and school editions can all be cataloged. When a book does have an ISBN, its check digit is verified on the phone.
How is it different from Z39.50 copy cataloging?
Z39.50 copies a record another library already made, so it finds nothing for books nobody has cataloged yet. CatalogLense does original cataloging from the book in your hand, and still uses a Library of Congress record whenever one exists.
Can records made with AI be trusted?
Every field shows where it came from, every value the AI read is looked for in a word-for-word transcript of your photos, and anything not printed on the pages is flagged. Nothing is saved to Koha until a cataloger has reviewed the record.
What else does the app do besides cataloging?
Catalog search, adding copies, editing records, shelf reading for inventory, checkout, renewals, check-in and holds at the circulation desk, barcode scanning, call numbers for spine labels, and reports on your collection.
How much does it cost?
Starter and Pro are free while CatalogLense is in beta. Free gives you 15 books to try it, Starter is for one cataloger with 100 books a month, Pro for a team of five with 500 a month, and Enterprise is arranged for library systems. Prices will be announced at launch — see Plans.
Where does my library's data go?
Catalog, patron and circulation data go only between the phone and your Koha: we never connect to your Koha, unless you have us host it under an Enterprise agreement. Book photos go through our server to the AI (Google or Anthropic) and aren't kept.
Can you host Koha for us?
Yes, on the Enterprise plan. We can set up and run Koha in the cloud for your library, propose other systems that work around Koha, and provide Koha plugins that connect them. Contact us to talk it through.
Contact
Curious? Ask us anything.
Whether it fits your library, a quote for Enterprise, or help getting your Koha ready — write to us. A person reads every message.
- Tell us your Koha version, if you know it.
- For Enterprise, roughly how many branches and catalogers.
Privacy
Your Koha never passes through us.
A small account
We keep your email, your library's name, its plan and how many AI credits it used. That's what the account is for.
Koha data stays in your Koha
Catalog, patron and circulation data go only to the server address you enter. We never connect to your Koha unless you have us host it, and Koha passwords stay in the phone's keychain.
Book photos go to the AI
Through our server to Google or Anthropic, and not kept. Accuracy reports tell us how well the AI read. They never contain photos or patron data. No ads, no tracking.