Learn from a few examples
Give 2–10 example texts per class and the service generalizes from them — no training required.
CatalogLatency Typically 1–3 seconds Async false Rate Limit Per API key Max Input Roughly a few hundred tokens per text
cn8/text-few-shot
Send a `text` plus an `examples` map — each key is a class label and each value is a
Learn more about Few-Shot Classification Text NLPsyncInstant previewtextanalysisnlp
Try it
Text to classify
Categories with examples
Result
Results show up here
The sample input is ready — just hit Run.
Overview
Send a text plus an examples map — each key is a class label and each value is a short list of example texts for that class. The service compares your text against those examples and returns the predicted label, a confidence value, and an all_scores map of every class to its score. Responses are synchronous, and there's no training step — change the classes any time by editing the examples map.
Key capabilities
Predicted label + all scores
Get the predicted label, a confidence value from 0 to 1, and a score for every class.
Change classes anytime
Add or remove classes whenever you like by editing the examples map — nothing is stored between calls.
Multilingual
Examples and input can be in any supported language, and you can mix languages.
When to use it
Custom taxonomy
Classify into domain-specific categories with a few examples each (e.g. ticket intents, product types).
Intent detection
Recognize user intents from a handful of example utterances per intent.
Evolving categories
Adjust your classes on the fly as your needs change.
Input & output
input
text: a string to classify, plus examples: a map of { label: [example1, example2, ...] }.
JSON body
output
The predicted label, a confidence value, and an all_scores map of every class to its score.
JSON
Guides & tips
How it works
- You provide a
textand anexamplesmap: each label points to a short list of example texts. - The service compares your text against the examples for each class and picks the closest one.
- The predicted class is returned at
data.label, its score asconfidence, and every class's score inall_scores.
Tips
- Send
examplesas an object —{label: [text, text, ...]}— not a list of label/text objects. - Use at least 2 examples per class, ideally 5–10, with varied phrasing for better accuracy.
- The top-level score field is
confidence— treat low values (for example below 0.5) as "no confident match" in your app. - The price is per request no matter how many classes or examples you send.
Specs
Schema
Request body
textrequiredstringThe text to classify.
examplesrequiredobject<string,array<string>>Map of label → list of example strings, e.g. {"food": ["...", "..."]}.
Response
labelstringPredicted label.
confidencenumberConfidence of the predicted label, from 0 to 1. (The field is named confidence.)
all_scoresobject<string,number>Map of every label to its score. Sort by value to rank the classes.
Pricing
Billed per request, regardless of how many classes or examples you send.
| Service | Unit | Price |
|---|---|---|
| Few-Shot Classification | item | $0.006/request |