Route user requests to specialized agents with GPT-4o Mini

This n8n workflow template is designed to route user input to specialized agents (like a Reminder Agent, Email Agent, etc.) using a structured output from a language model. Here's a complete description of what it does and how each part works:

🔁 Workflow Purpose:

This template receives a user's request via Webhook, processes it using an LLM, extracts structured data like the agent name and user query, and routes the input to the appropriate sub-workflow (agent) based on the specified agent type.

🧩 Workflow Breakdown:

1. Webhook (Trigger)

• Node: Webhook
• Purpose: Accepts a POST request from any frontend or API source. It contains the raw user input.

2. GPT Model (LLM Inference)

• Node: GPT 4o Mini

• Purpose: Interprets the user input and determines:

  • Which agent should handle it (e.g., "Reminder Agent", "Email Agent", etc.)
  • The actual user request (in structured format)

3. Auto-Fixing Output Parser

• Node: Auto-fixing Output Parser
• Purpose: Ensures that the output from the LLM matches the expected structure. If there's a mismatch, it automatically corrects it using a re-prompt.

4. Structured Output Parser

• Node: Structured Output Parser

• Purpose: Converts the language model's response into a strict JSON structure with keys like:

  • "Agent Name"
  • "user input"
  • "sessionID"

5. Agent Router

• Node: Switch ("Agent Route")

• Purpose: Based on "Agent Name", it routes the input to one of the following sub-workflows:

  • 📅 Reminder Agent
  • 📧 Email Agent
  • 🧾 Document Agent
  • 🤝 Meeting Agent

6. Sub-Workflow Call (Execute Workflow)

Each agent is implemented as a separate n8n workflow:

• The input is forwarded to the selected agent.
• For example, if "Agent Name" is "Reminder Agent", the workflow "Reminder Agent" is called with "user input".

7. Webhook Response

• After the sub-agent workflow finishes, a Respond to Webhook node sends the output back as an HTTP response.

✅ Key Features:

• Fully modular and extensible
• LLM-driven routing using OpenRouter GPT-4o
• Auto-corrects structured output errors
• Clean separation of concerns (agent logic is decoupled in sub-workflows)
• Easily add more agents by updating the switch logic

📦 Use Case Examples:

• User says: “Remind me to call my mom tomorrow.” → Routed to Reminder Agent

• User says: “Send an email to the HR team.” → Routed to Email Agent

• User says: “Schedule a meeting with John next week.” → Routed to Meeting Agent