kayetanaSort text without reading every line.
Support tickets, emails, feedback: kayetana assigns each text one or more labels and tells you how confident it is. A few examples are all it takes.
Examples in, labels out.
Write a few examples for each label, say “Account,” “Shipping” and “Billing.” Then enter new texts. kayetana assigns each one a label and shows its confidence as a percentage.
- Examples right in the request
- Confidence per label
- One or more labels per text
With examples, or with your own classifier.
| With examples | Trained classifier | |
|---|---|---|
| Setup | A few examples per label in the request | Labeled dataset, trained once |
| Best for | Quick tests, changing categories | Fixed categories, high volume |
| One or more labels per text | ✓ | ✓ |
| Confidence per label | ✓ | ✓ |
| Available via API | ✓ | ✓ |
Where classification takes work off your plate.
Support tickets
Sort new tickets by topic and send them straight to the right team.
Shared inbox
Split emails to your general address into inquiries, applications, invoices and complaints.
Feedback
Tag customer feedback as positive, neutral or critical.
Contract clauses
Map clauses to topics like liability, term or data protection.
Multiple topics
An email about shipping and billing? Texts can carry more than one label.
Review where it counts
Confidence per label shows which cases a person should look at.
How to build your own sorting.
Test with examples
Use the Playground to check whether your categories separate cleanly.
Create a dataset
Upload labeled texts, with one or more labels per text.
Train a classifier
Under Fine-tuning, turn it into a classifier “clf-name” built on an embedding model.
Use it via API
Your app sends texts to classify and gets back labels and confidence.
Sort your tickets automatically.
Bring a few examples from your inbox, and we'll show you the results in the demo.
Frequently asked questions
How many examples do I need?
For the example-based approach, a few per label are enough. For your own classifier, you create a dataset of labeled texts.
Can one text get more than one label?
Yes. Texts can carry one or more labels, and kayetana shows the confidence for each.
Which model does the work?
Classification runs on an embedding model, and bge-m3 comes included. Trained classifiers are named “clf-name” and are built on an embedding model.
How do I connect it to our ticketing system?
Through the classify endpoint of the REST API. Your system sends the texts and gets labels and confidence back.