How the platform works

Transkribus is one platform: the app and the APIs operate on the same objects (collections, documents, models, datasets, sites), and this documentation is the developer view of it. Text recognition happens to be the first endpoints in the roll-out; further API surfaces, such as sites and search, join this documentation as they ship and are announced in the changelog.

The first surface is the job API, and its core is one loop you can run with nothing but an account and an image:

submit an image + a model ID → get a job → poll → fetch the result

The image travels inside the request (base64 or URL), the model ID selects what runs, and the job contract (statuses, polling, results, errors) is identical for every job type. It is described once in How jobs work.

How it connects to the Transkribus app

The resources you use in the app are the same objects; models you train in Transkribus are already usable here by ID. Capabilities documented as working in the app today get their API surfaces announced in the changelog as they ship.