Text Classifier

Classify text into categories using an LLM.

Written By pvdyck

Last updated About 5 hours ago

Text Classifier

The Text Classifier uses an LLM to assign input text to one or more predefined categories. Each category becomes a separate output, enabling workflow branching based on classification.

How It Works

You define a list of categories (with optional descriptions). The node sends the input text to the LLM, which classifies it into one of the categories. Each category maps to a separate output connector, so you can wire different downstream logic per category.

Parameters

ParameterDescription
TextThe input text to classify. Supports expressions like {{ $json.message }}.
CategoriesList of classification categories. Each has a Name and optional Description to guide the LLM.
System PromptOptional instructions for classification behavior.
Allow MultipleWhen enabled, the model can assign the text to more than one category. Default: single category.
When No MatchControls what happens when the text doesn't match any category. Discard Item (default): silently drops the item. Output on Extra "Other" Branch: routes unmatched items to a separate "Other" output connector.
Enable Auto-FixingRe-prompts the model if its output doesn't match a valid category, sending the parsing error back to the LLM for correction.

Category Configuration

Each category requires:

  • Name — The label (e.g., "Sales", "Support", "Billing").
  • Description — (Optional) Helps the LLM understand what belongs in this category.

The node creates one output per category, enabling direct routing in your workflow.

Sub-Node Connections

InputRequiredDescription
AI Language ModelYesThe LLM powering the classification.
Output ParserNoAdditional output formatting.

Example Use Cases

  • Route support tickets to Sales / Support / Billing teams
  • Classify emails by urgency (High / Medium / Low)
  • Tag content by topic (News / Tutorial / Opinion)
  • Sort feedback by product area

Tips

  • Write clear category descriptions — "Billing: questions about invoices, payments, refunds" works better than just "Billing".
  • Use Allow Multiple when text may belong to several categories (e.g., a ticket about both billing and a bug).
  • Connect each output to a different downstream workflow branch for automatic routing.
  • For simple positive/negative/neutral classification, consider the Sentiment Analysis node instead.

Related