Search...
  • Pricing
Get Started

Start typing to search across models, sites, and blog posts...

Transkribus API · Generally available

Transkribus, from your own code

The Transkribus API runs the same AI as the Transkribus app: Text Titan, 300+ public models, Smart Extract, and every model you train yourself. Send a page image, get back text, layout and structured data.

Read the documentationSee plans

Scholar plan and above · API jobs at half the credits of the app · Processed on our own servers in Austria

submit.sh
curl -X POST https://api.transkribus.org/v2/processes \
  -H "Authorization: Bearer $ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "config": { "textRecognition": { "htrId": 38230 } },
    "image": { "imageUrl": "https://example.org/scan.jpg" }
  }'

Get a token. Then three requests.

Sign in with your Transkribus account, submit a page, poll the job, download the result. Every job type follows the same contract.

1

Authenticate

Exchange your Transkribus credentials for an access token via OpenID Connect.

POST /openid-connect/token
2

Submit a page

Send one image as base64 or URL, with the ID of the model to run.

POST /v2/processes
3

Poll the job

Poll on an interval, or long-poll and get an answer the moment the status changes.

GET /v2/processes/longpoll/{id}
4

Download the result

Take the JSON from the job, or fetch PAGE XML, ALTO XML or a ZIP archive.

GET /v2/processes/{id}/page

Included from the Scholar plan

The Transkribus API comes with every plan from Scholar upwards. You sign in with your Transkribus account, and API jobs use credits from your plan at half the rate of the same job in the app.

Scholar

For individual researchers: one seat, Text Titan II and the advanced AI tools, plus API access.

Team

Five seats and a shared credit pool for your team, plus API access.

Organisation

15+ seats, a dedicated success manager and custom onboarding, plus API access.

Compare plans

One API, two ways to read a page

Pick the job type in the request. Polling, results and errors work the same for both.

Text recognition

Turn printed and handwritten pages into text with full layout: regions, baselines, lines and reading order. Start with Text Titan, our super model for mixed print, cursive and Kurrent, or pick one of 300+ public models for a specific script, language or century. Models you train in Transkribus are addressed exactly the same way.

Super models work across hands and languages without training
Small trained models run faster and cheaper on one hand or collection
Your own models, available by ID as soon as training finishes
text-recognition.json
{
  "config": {
    "textRecognition": { "htrId": 38230 }
  },
  "image": {
    "imageUrl": "https://example.org/scan.jpg"
  }
}

Growing into the full Transkribus API

Text recognition and Smart Extract are live. The other capabilities of the Transkribus app follow one by one, and each new surface is announced in the developer changelog.

Live nowIn the app today, on the API next

Live: Text recognition

Print and handwriting with layout, using super models, public models or your own.

Live: Smart Extract

Structured data from a page image in one job: text, tables, entities and page classes.

Live: JSON, PAGE XML, ALTO

Results straight from the job, in the format your systems already read.

Next: Layout, tables, fields

Layout analysis, table recognition and field models as jobs of their own.

Next: Model training

Train a model from your ground truth and use it by ID, all from code.

Next: Documents, export, Sites

Collections and documents, exports in every format, and publishing to Transkribus Sites.

Follow the changelog

The Transkribus advantage

Train in the app. Run it through the API.

Your domain experts create ground truth and train models in Transkribus, without writing code. Your developers run those models at scale through the API. One platform, two workflows, and the model moves between them by ID.
Ground truth in a visual editor
Text recognition models trained in a few clicks
Trained models are usable through the API right away
300+ public models when you don't want to train
Browse public models
Training a model in Transkribus

Developer documentation

Everything else is at docs.transkribus.org

Guides for authentication, your first job, the job lifecycle, errors, credits and limits, plus the complete API reference with every endpoint, schema and example. When a new API surface ships, it lands there first.
Step-by-step guides from token to result
Interactive API reference generated from the OpenAPI spec
Changelog for every new endpoint
Open the documentation
The Transkribus developer documentation

Case study

FromThePage: AI-assisted crowdsourcing via API

FromThePage integrated the Transkribus API to add an AI Assist feature to their crowdsourcing platform. Volunteers see AI-generated transcriptions next to the document image, so they transcribe faster and humans stay in the loop. Used by Harvard, Stanford, the British Library and dozens of heritage institutions.
AI pre-transcription speeds up volunteer work
Volunteers correct, the model does the first pass
Used by major research institutions worldwide
Read the case study
FromThePage AI Assist powered by the Transkribus API

Try it yourself

Upload a document, printed or handwritten, and see the API response in real time.

Drag an image here

Select a file...

PNG or JPEG, max 10 MB

Transkribus API: frequently asked questions

Anyone with a Transkribus account on the Scholar plan or above. You sign in with the same credentials you use in the app. See plans and pricing.

API jobs use credits from your Transkribus account, at half the credits the same job costs in the Transkribus app. For volume workloads and custom terms, talk to our team.

All public models, including the Text Titan super model, and any model you have access to in Transkribus, including the ones you train yourself. You address every model by its ID.

JPEG, PNG and TIFF up to 20 MB, one image per job, sent as base64 or as a public URL. Results come as JSON in the job response, as PAGE XML, as ALTO XML or as a ZIP archive, and stay available for 24 hours after the job finishes.

On our own servers in Austria. The image you send is used for that job only and is not stored, and results are removed 24 hours after the job finishes.

Yes. For processing that never leaves your network, see Transkribus On-Premises.

metagrapho was the earlier name of the Transkribus API. It is now simply the Transkribus API, documented at docs.transkribus.org.
200M+ pages processed300+ AI models500,000+ usersHosted in Austria

Built on trust, hosted in Europe.

Your documents are processed on our own servers in Austria. No US cloud, no Big Tech dependency.

Processed in Austria

Images are used for the job only and not stored. Results are removed 24 hours after the job finishes.

Production scale

The same infrastructure that has processed 200M+ pages in Transkribus. On-premises deployment is available when documents must stay in your network.

Cooperative, not corporate

READ-COOP is a European cooperative owned by its members. No vendor lock-in, no exit strategy.

Start building with the Transkribus API

Get a Scholar plan or above, request a token and send your first page. For volume workloads and custom terms, talk to our team.

Read the documentationTalk to our team
RESTJSON over HTTPS
300+AI models
Austriaown servers