Route AI tasks between OpenAI agents with confidence-based email fallback

Overview

This workflow demonstrates an AI task routing system using multiple agents in n8n. It analyzes incoming user requests, determines their complexity, and routes them to the most appropriate AI agent for processing.

A Supervisor Agent evaluates each request and classifies it as either simple or complex, returning a confidence score and reasoning. Based on this classification, an orchestrator agent delegates the task to the correct specialized agent.

The workflow also includes a confidence validation mechanism. If the classification confidence falls below a defined threshold, an email alert is sent to an administrator for manual review.

This architecture helps build scalable AI systems where tasks are intelligently routed to agents optimized for different levels of complexity.

How It Works

1. Webhook Trigger
   The workflow starts when a request is received through a webhook endpoint.

2. Workflow Configuration
   The request and a configurable confidence threshold are stored using a Set node.

3. Supervisor Agent Classification
   The Supervisor Agent analyzes the user request and determines whether the task is simple or complex, returning a confidence score and reasoning.

4. Structured Output Parsing
   The classification result is parsed using a structured output parser to ensure reliable JSON formatting.

5. Confidence Validation
   An IF node checks whether the confidence score meets the configured threshold.

6. Agent Orchestration
   If the confidence is sufficient, an orchestrator agent delegates the task to either:

   • Simple Task Agent for straightforward questions
   • Complex Task Agent for tasks requiring deeper reasoning

7. Fallback Handling
   If the confidence score is too low, the workflow sends an email alert requesting manual review.

8. Webhook Response
   The final AI response is returned to the original requester through the Respond to Webhook node.

Setup Instructions

1. Add OpenAI credentials to all OpenAI model nodes:

   • Supervisor Model
   • Executor Model
   • Simple Agent Model
   • Complex Agent Model

2. Configure the Workflow Configuration node:

   • Set the userRequest placeholder if testing manually.
   • Adjust the confidenceThreshold if required.

3. Configure the Email Send node:

   • Enter sender and administrator email addresses.
   • Connect SMTP or your preferred email credentials.

4. Activate the workflow and send requests to the Webhook endpoint to start task processing.

Use Cases

• AI support systems that route queries based on complexity
• Customer service automation with intelligent escalation
• Multi-agent AI architectures for research or analysis tasks
• AI workflow orchestration for automation platforms
• Intelligent request classification and routing systems

Requirements

• OpenAI API credentials
• Email (SMTP) credentials for alert notifications
• A system capable of sending requests to the workflow webhook