💬 Chatbot that Knows Your Documents (OpenRouter + Product Management Agent)

🔀 Workflow: Chatbot that Knows Your Documents

This blog contains the Webhook + Retrieval-Augmented Generation (RAG) workflow for n8n, powered by OpenRouter’s deepseek-chat model. It demonstrates how prompt wording and topic specificity change the AI’s response — while adding a RAG tool for Product Management insights.

---

✨ Overview

This workflow expands the 201-basic pattern by introducing memory, OpenRouter free-tier LLM, and a RAG API tool for contextual answers. It creates a Product Management AI Agent that:

• 🧠 Uses RAG for Product Management questions
• 💬 Uses the LLM (deepseek-chat) for general or AWS-related questions
• 🙅 Politely declines out-of-scope or non-relevant questions
• 🧍 Keeps short, simple replies (2–3 sentences), emphasizing clarity and prompt impact

---

🔄 How It Works

graph LR
  A["🌐 Webhook (POST)"] --> B["🧠 AI Agent"]
  B --> C["🤖 OpenRouter Chat Model (deepseek-chat)"]
  B --> D["🛠️ RAG Tool (Product Management API)"]
  B <--> E["🗂️ Memory (sessionKey = username)"]
  B --> F["↩️ Respond to Webhook"]

1. Webhook receives JSON (query, username).

2. AI Agent decides how to respond:

3. If the topic involves Product Management → call the RAG Tool.

4. Otherwise → use the deepseek-chat model directly.

5. For vague questions → respond generally and guide users to be more specific.

6. Memory Buffer (Window) stores short-term conversation context by username.

7. Response Node returns the generated reply to the client.

---

🏛️ Architecture

---

🛂 Inputs (JSON Body)

Field	Type	Required	Description
query	string	✅	User’s input message
username	string	🟢 Recommended	Session key for conversational memory

Example

{
  "query": "What are best practices for Product Managers when prioritizing features?",
  "username": "demo-user-1"
}

---

📤 Output

• HTTP 200 OK with a short AI reply.

• Reply style examples:

• “(Answer from Product Management knowledge base)” — via RAG Tool

• “(General or AWS-based answer)” — via model
• “(Out-of-scope polite refusal)” — for unrelated queries

---

⚙️ Setup

1. Import 301-webhook-rag.json into your n8n Cloud workspace.

2. Credentials → Configure:

3. 🔑 OpenRouter API (create a free account at openrouter.ai)

4. 🔑 Optional: API key for the Traversaal Product Docs API (used by the RAG Tool)

5. In the “RAG Tool” node, replace:

bash Authorization: your_traversaal_api_auth_brearer_token_here

with your actual bearer token.

1. Activate the workflow and copy the Production Webhook URL (from the Webhook node).
2. Optionally adjust the systemMessage inside the AI Agent node to modify tone or domain focus.

---

🧪 Try It

Option A — Google Colab (Recommended)

1. Open the instructor’s Colab notebook: 301 Webhook + RAG — Client (Colab)
2. Click File → Save a copy in Drive.
3. Replace WEBHOOK_URL with your Production Webhook URL from n8n.

4. Run all cells and test:

5. 🧠 “What are best practices for Product Managers?”

6. ☁️ “Tell me about AWS S3 buckets.”
7. 🏙️ “What is the capital of France?”

💡 Use the same username to maintain context; new names start fresh sessions.

---

Option B — cURL

WEBHOOK_URL="https://<your-n8n>/webhook/<id>"   # Use Production URL
curl -X POST "$WEBHOOK_URL" \
  -H "Content-Type: application/json" \
  -d '{"query":"List key principles of Product Management","username":"demo-user-1"}'

Option C — Postman

1. Create a POST request → Production Webhook URL
2. Body → Raw → JSON:

json { "query": "How can I improve sprint planning as a Product Manager?", "username": "demo-user-1" }

1. Click Send → view AI reply.

---

🧠 What to Notice (Teaching Points)

• Demonstrates how prompt wording changes the model’s specificity.
• Shows AI tool routing (LLM vs RAG) within a single workflow.
• Highlights the benefit of contextual memory via sessionKey.
• Provides a budget-friendly OpenRouter setup (deepseek-chat is free tier).
• Encourages experimentation with tone and system instructions.

---

📚 References

• 📖 Traversaal Product Management Docs API
• 📖 OpenRouter — deepseek-chat model
• 📖 n8n — AI Agent node
• 📖 n8n — HTTP Request Tool
• 📖 n8n — Memory Buffer Window

---

🪄 Learn how to chain OpenRouter, RAG, and custom APIs in your own workflows — the same pattern used here.

---

Don’t forget to check out my Agentic AI System Design for PMs course on Maven if you are interested to be a part of something bigger.

---

➡️ Next