--- title: "Text Classifier - Private AI Categories | CanDoYa" description: "Use a private AI text classifier to sort English text by topic, message type, or meeting-note purpose. Runs on your device with no text upload." url: https://candoya.com/tools/text-classifier/ locale: en type: browser-tool --- # Free Text Classifier Category: Social & Creator ## What is a text classifier? A text classifier assigns a passage to one category from a defined set. This private tool ranks English text by content topic, message type, or meeting-note purpose using an on-device MobileBERT model. You can inspect every category score, while the text itself stays inside your browser. **This text classification tool uses Xenova's MobileBERT MNLI model through Transformers.js.** Instead of asking you to invent labels, it provides three practical five-label sets. Content topic covers world news, business, science and technology, sports, and culture. Message type separates questions, requests, complaints, feedback, and status updates. Meeting note identifies action items, decisions, risks, open questions, and background information. The worker downloads public model files only after you start a classification. It tries WebGPU first and falls back to WebAssembly on the CPU. The first download is roughly 22 to 28 MB, depending on the backend, and browsers normally cache it. Your entered text is not included in those file requests. ## How to use it 1. **Choose a classification set** - Select Content topic, Message type, or Meeting note. The five labels in that set appear below the selector. 2. **Enter a focused passage** - Paste an English sentence or short paragraph. Keep mixed-topic documents separate so one passage represents one main idea. 3. **Run the on-device model** - Select Classify text and allow the first model download to finish. WebGPU is used when available, with a CPU fallback. 4. **Review the full ranking** - Compare the leading category with the remaining scores, note any close-match warning, and copy the result if it is useful. ## Who it's for - **Editors and researchers** can triage short English articles into a consistent topic set before reviewing edge cases by hand. - **Support and operations teams** can distinguish questions, requests, complaints, feedback, and status updates in sample messages. - **Meeting and voice-note users** can check whether a transcript excerpt reads like an action item, decision, risk, open question, or background note. - **Developers and analysts** can test a ready-made classification workflow before investing in training data or a hosted API. ## Expert note This classifier compares the same passage with every label as a natural-language inference problem. Scores are relative to the selected set: changing from Content topic to Message type changes both the question asked of the model and the meaning of every score. Validate representative examples before using the ranking in a workflow. ## Frequently asked questions ### Is my text uploaded to a server? No. Classification runs in a worker inside your browser, and your pasted text is not sent to CanDoYa, jsDelivr, or Hugging Face. The worker requests the public Transformers.js runtime and model files. Those download requests do not contain the text you entered. ### Is the text classifier free? Yes. There is no account, API key, payment, or server-side usage quota. Your device performs the inference. The practical costs are the first model download, local processing time, browser storage, and any network data used to fetch the public model files. ### How much text can I classify? The tool accepts any pasted text but classifies the first 1,600 Unicode characters after whitespace is normalized. One focused sentence or short paragraph usually works better than a long document with several unrelated topics, requests, or decisions competing for one label. ### What categories does the classifier use? You can choose one of three fixed sets. Content topic has five broad editorial topics. Message type has question, request, complaint, feedback, and status update. Meeting note has action item, decision, risk, open question, and background information. Every run ranks all five labels in the chosen set. ### How accurate is the AI text classifier? Accuracy depends on how closely your passage and labels resemble the English examples the MobileBERT model learned from. A wide score gap is more persuasive than a narrow one, but neither is proof. Short fragments, mixed topics, sarcasm, specialist language, and ambiguous intent can all produce weak rankings. ### Which languages does the classifier support? The MobileBERT natural language inference model is trained for English, so this page marks the results English only. Other languages may still produce scores, but those values do not have the same training basis and should not be treated as reliable classifications. ### Does the classifier work without WebGPU? Yes. The worker tries WebGPU when a compatible browser and graphics adapter expose it. If WebGPU is unavailable or fails, the tool reloads the same model with the WebAssembly CPU backend. That fallback keeps text local but can take longer to load and run. ### How is this different from a zero-shot classifier? This page gives you three ready-made label sets for common tasks, so you can classify text immediately and compare consistent categories. The zero-shot classifier is the flexible option when you need to write your own labels or decide whether several custom labels can apply at once. ## Related tools - [Zero-Shot Classifier](https://candoya.com/tools/zero-shot-classifier/) - [Sentiment Analysis](https://candoya.com/tools/sentiment-analysis/) - [Text Summarizer](https://candoya.com/tools/text-summarizer/) - [Keyword Extractor](https://candoya.com/tools/keyword-extractor/) - [Language Detector](https://candoya.com/tools/language-detector/) --- *Markdown edition of , published for AI agents and other automated readers. Index of key pages: .*