Skip to content
Language:DeutschEnglishFrançaisItaliano Last updated: 2026-08-18

AI-assisted cataloguing

Anton can have suggestions for cataloguing generated by a language model: it reads the attached media and proposes titles, text fields, events, keywords, actors, places and languages.

Has to be enabled first

AI-assisted cataloguing is currently not in operation by default. It requires two things: a release in the server configuration, which is not reachable in the admin area — with Anton as a Service, k & r sets it — and the switch in the archive itself. In addition, a cost limit has to be stored; if it is missing, all requests are aborted.

Generating suggestions

The edit form contains the AI cataloguing block; the button of the same name at the top jumps to it. A click on generate AI suggestions sends the request; the suggestions then appear as chips next to the fields concerned, each with apply, append and ignore.

Nothing is written to the database automatically in the process — the suggestions land in the form and are only saved with the normal save button. Generating them requires the editor role.

What Anton adopts on its own — and what it does not

The defaults are deliberately unequal:

Suggestion Default
Actor, place, keyword — already present is linked
Actor, place, keyword — to be newly created is discarded
Titles and text fields are discarded

That means: the AI never changes the title on its own. Anyone who wants a suggestion has to apply it explicitly. And new authority records do not come about incidentally — the threshold for new actors and keywords deliberately remains high.

Anton creates suggested events as creation date. If it is a different event type, it has to be corrected by hand after being applied.

Data protection

Anton marks every AI profile according to where it processes the data: 🇨🇭 for models hosted in Switzerland, ⚠️ for all others. With a non-Swiss profile, a red warning appears:

⚠️ Caution — data protection This profile processes data outside Switzerland and is NOT FADP/GDPR compliant. Use it exclusively for data that you would also publish openly.

A Swiss profile is preset. Choosing a different one is a deliberate decision in each individual case: the selection is collapsed, applies only to the current request and is not remembered.

Check before sending

With the request, the contents of the unit of description — media included — leave the archive. For blocked holdings, personal data and everything subject to a protection period, this is not permissible with a non-Swiss profile.

Every request is logged; consumption, costs and an audit trail are available under Admin → AI cataloguing.