Text Classifier

Put a text into one of the categories you list, with a chat model or on-device.

Try it

Workflow · classify-and-route.awf 8 Schritte

Classify and Route: ein Workflow mit 8 Schritten, beginnend mit Run now und endend mit Other branch, mit 1 Verzweigung.

1. 1 Run now(When Started – zur Dokumentationsseite) When Started
2. 2 Message to sort(Edit Fields – zur Dokumentationsseite) Edit Fields
3. 3 Classify the message(Text Classifier – zur Dokumentationsseite) Text Classifier + Local AI model
4. 4 Route by category(Switch – zur Dokumentationsseite) Switch
5. • 5 Billing branch(Display Markdown – zur Dokumentationsseite) Display Markdown wenn Billing
   • 5 Technical branch(Display Markdown – zur Dokumentationsseite) Display Markdown wenn Technical
   • 5 Sales branch(Display Markdown – zur Dokumentationsseite) Display Markdown wenn Sales
   • 5 Other branch(Display Markdown – zur Dokumentationsseite) Display Markdown wenn andernfalls

In AWFlow öffnen .awf kopieren

What it does

Text Classifier reads Text and returns the best matching category from Categories in category. What you connect to Model decides how; there is no setting to switch:

• A chat model (for example Chat OpenAI or Web LLM) is told to answer with exactly one category name and nothing else, or unknown when none fits.
• A Local Classifier runs in your browser: free, private and offline. With a zero-shot model, your Categories are scored and the best one wins. With a fixed-label model (sentiment, toxicity), the model’s own labels are used and Categories is ignored. System message does not apply.

The node has one output. To send each category down its own path, follow it with a Switch.

When to use it

• Sort support requests into billing, technical, sales and route each to the right team.
• Tag articles or emails by topic before you save them.
• Detect the intent of a chat message (question, complaint, order).

To label tone only, use Sentiment Analysis. To classify on-device without a model node, use Classify Text.

Inputs and settings

Einstellung	Typ	Was es bewirkt
Text PFLICHT(erforderlich)	Long text	The text to classify.
Categories PFLICHT(erforderlich)	Long text	Comma-separated category names to classify the text into.
Options	Group	The options for the node.
Options › System message	Long text	The system message to use.

Fill Text with an expression such as {{ $input.text }}. Write Categories as a comma-separated list, for example billing, technical, sales. Short, clearly different names give the most reliable results. Use System message to describe a category when its name alone is ambiguous, such as "sales" means questions about buying or pricing.

Outputs

• Output

Each output item has these fields:

Field	Typ	What it holds
category	Any	The chosen category
scores	Key-value pairs	Every label's score, on-device

Shape of an output item

{
  "category": null,
  "scores": {}
}

category is one of your category names, or unknown when none fits (or when the on-device model returns no label). scores is only there with a Local Classifier.

Dependencies and credentials

Slot	Accepts	Connections
Model PFLICHT(erforderlich)	chat model, local classifier	One node.

Connect a chat model, which holds its own API key, or a Local Classifier with a zero-shot model to classify on-device with your own categories.

Example workflow

1. Get Selected Text reads a support email you highlighted, with Output field name text.
2. Text Classifier with Text {{ $input.text }} and Categories billing, technical, sales, with an Chat OpenAI model on Model.
3. Switch in Rules mode with one rule per category, each checking that {{ $input.category }} equals the category name, and Add A Fallback Branch on for unknown.
4. Each branch posts to a different Slack channel.

Common issues

• category is often unknown. Your categories do not cover the texts you receive, or the names are vague. Add a category, or explain them in System message.
• The category does not match your Switch rules. A chat model may change the capitalisation or add words. Add Answer with the category name exactly as written. in System message.
• “Add at least one category: the zero-shot classifier scores the text against them.” A zero-shot Local Classifier is connected and Categories is empty.
• “Model not found”. The connected chat model node is not set up: open it and choose a model and its credential.

Related nodes

• Switch: send each category down its own path.
• Sentiment Analysis: label tone.
• Classify Text: on-device classification without a model node.
• Local Classifier: the on-device model for this node.