Router Block

The Router block splits workflow execution into up to 4 different paths based on AI-powered classification. Unlike the Condition block (which only does True/False), the Router can direct flow to multiple named destinations — making it ideal for classification, categorization, and multi-branch decision trees.

Overview

When to Use

• You need 3 or more branching paths (Condition only supports 2)
• You want AI to classify input into categories
• You're building a multi-intent chatbot or ticket routing system
• You need to route based on content type, language, or topic

Configuration

Routes

Define 2–4 named routes. Each route has a label that the LLM uses to make its decision. The LLM responds with the route name, and the workflow follows that path.

Outputs

Example: Customer Support Ticket Router

Goal: Automatically route support tickets to the right department.

Workflow:

[Starter] → [Router] → [Agent: Billing]     (Route 1)
                      → [Agent: Technical]   (Route 2)
                      → [Agent: Sales]       (Route 3)
                      → [Agent: General]     (Route 4)

Configuration:

• Route Prompt: Classify this customer support message into the correct department. Consider the topic, keywords, and intent.
• Context: {{starter.input}}
• Model: gpt-4o-mini
• Routes: Billing, Technical, Sales, General

How it works:

1. Customer submits a support message
2. Router sends it to GPT-4o-mini for classification
3. Model returns one of the 4 route names
4. Workflow branches to the specialized Agent for that department
5. Each department Agent has its own system prompt and tools

Tips

• Use descriptive route names — the LLM uses them to decide, so Billing Support is better than Route1
• Keep to 2–4 routes — more routes reduce classification accuracy
• Use Router over chained Conditions when you have 3+ paths — it's cleaner and uses only 1 LLM call
• GPT-4o-mini works great for simple classification — save expensive models for reasoning tasks