Tools

Extend your Agent node's capabilities with specialized tools for accurate task execution.

Overview

Tools are specialized capabilities you can attach to Agent nodes. They allow the agent to perform specific tasks with higher accuracy than relying solely on LLM prompting. Tools are particularly useful for classification, data extraction, and other structured tasks.

Why Use Tools?

Higher Accuracy

Pre-trained models are fine-tuned for specific tasks, providing more consistent and accurate results than general LLM prompting.

Token Savings

Tools handle specific tasks efficiently without consuming LLM tokens for classification or extraction logic.

Available Tools

IT Services Intent Classifier
Pre-trained intent classification for IT service company use cases

This tool is trained and fine-tuned using distilbert-base-uncased, a lightweight but powerful transformer model. It accurately classifies user queries into predefined intent categories.

Supported Intents

Requirement Submission

User is submitting a project requirement or RFP

General Query

General questions about services or company

Contact Details

User is asking for contact information

Feedback Submission

User is providing feedback about services

Appreciation

User is expressing thanks or positive sentiment

Greeting

User is saying hello or starting conversation

Job Application Submission

User is inquiring about jobs or submitting an application

Use Case: Mixed Inquiry Routing

IT service companies often receive both client inquiries and job applications through the same channels. This tool helps automatically separate them with 95% accuracy, ensuring leads go to sales and applications go to HR.

How to Attach a Tool

  1. Open your workflow and select an Agent Node
  2. Navigate to the Tools section in the node configuration
  3. Click Add Tool and select from available tools
  4. Configure any tool-specific settings if required
  5. Save your changes

Once attached, the agent will automatically use the tool when processing relevant queries. You can reference the tool's output in Condition nodes to route conversations based on the classified intent.

LLM vs Tool Classification

AspectLLM PromptingIntent Classifier Tool
AccuracyVariable, can hallucinateConsistent, ~95% accuracy
Token UsageConsumes LLM tokensNo LLM tokens used
SpeedDepends on LLM response timeFast, dedicated model
FlexibilityAny classification taskPre-defined intents

Coming Soon

Custom Intent Classifiers

We're working on the ability to train custom intent classifiers for your specific use cases. Define your own intents and train models on your data.

Next Steps

Condition Node

Route conversations based on tool outputs