Boneyard Tools

Zero-Shot Text Classifier

Sort text into your own categories without training a thing. Type the candidate labels you care about, such as billing, bug report, or feature request, and a BART-large-MNLI AI model scores how well your text matches each one. It is perfect for tagging support tickets, emails, or feedback. The model runs entirely in your browser, so nothing is uploaded. It is a large model, around 270 MB, that downloads once on first use, then is cached.

How to classify text with custom labels

  1. Paste the content you want to classify into the text box.
  2. Enter your candidate labels, separated by commas, and pick single or multi-label mode.
  3. Click Classify and read the ranked labels with their match percentages.

Examples

Triaging a support message

Text: My invoice charged me twice this month. Labels: billing, bug report, feature request
billing 96%, bug report 3%, feature request 1%

Frequently asked questions

Is my text uploaded anywhere?

No. Nothing is uploaded. The BART-large-MNLI model runs entirely in your browser via WebAssembly, so your text and labels stay on your device and are never sent to a server. Only the model itself is downloaded, once, then cached.

What does zero-shot classification mean?

Zero-shot means the model can classify text into labels it was never specifically trained on. You invent the categories on the spot and the model scores how well the text fits each one, with no training step or examples required.

What is the difference between single and multi-label mode?

In single-label mode the scores compete and add up to 100%, so you get the one best category. In multi-label mode each label is scored independently, so a piece of text can match several categories at once. Use multi-label when more than one tag can apply.

Which AI model does this use?

It uses BART-large-MNLI, a model fine-tuned on natural language inference that is the standard choice for zero-shot text classification. It runs locally in your browser through transformers.js and ONNX.

Why is the first run slower than other tools?

BART-large-MNLI is a larger model, around 270 MB, so the first download takes longer than smaller text tools. Once your browser caches it, later runs are much faster and work offline.

Related tools