Local Classifier
Local Classifier
Section titled “Local Classifier”An on-device text classifier you plug into the Model port of the Text Classifier or Sentiment Analysis agent. With it connected, the agent classifies locally in your browser instead of prompting a chat model — free, private, and offline.
What it does
Section titled “What it does”Exposes an installed classification model as a dependency. The agent it’s connected to decides how to use it:
- Zero-shot models (e.g. DeBERTa zero-shot) score the text against the agent’s own labels — Text Classifier’s Categories, or Sentiment Analysis’s Custom categories (else
positive, negative, neutral). - Fixed-label models (e.g. a sentiment or toxicity classifier) use their own labels. For Sentiment Analysis, common sentiment labels (
POSITIVE/NEGATIVE,LABEL_0..2,1 star–5 stars) are mapped ontopositive/negative/neutral.
What you connect decides the engine — there is no toggle: a chat model runs the LLM path, a Local Classifier runs on-device.
When to use it
Section titled “When to use it”- Route or tag text in a workflow that must stay private or run offline.
- Classify high volumes of text without paying per token.
- You don’t need an explanation — just a label and scores.
Inputs and settings
Section titled “Inputs and settings”| Setting | Notes |
|---|---|
| Model | An on-device text classifier: zero-shot (uses your labels), sentiment, or toxicity. Pick an installed one, or download it from the picker. Cleared when a workflow is published, so each user picks their own. |
Outputs
Section titled “Outputs”Returns a classifier dependency for the agent’s Model port. The agent’s output gains a scores map ({ label: score }) alongside its usual result — see the agent pages.
Troubleshooting
Section titled “Troubleshooting”- “Choose an on-device classifier in the Local Classifier node.” — the Model field is empty.
- “Model … isn’t available. Install it from Local AI.” — install the model from the Local AI page, or download it from the picker.
- The agent ignores my categories — the model is a fixed-label classifier. Use a zero-shot model to classify into your own labels.
- “Add at least one category” — Text Classifier with a zero-shot model needs at least one category.
Related pages
Section titled “Related pages”- Text Classifier, Sentiment Analysis — agents that accept it.
- Classify Text — standalone zero-shot classification node.
- Local Q&A Model — the on-device model for the Q&A Agent.
- Local AI — install and manage on-device models.