1Manual Trigger → AI Agent → Output

1You are a helpful assistant that can answer questions and perform tasks.
2When asked about weather, use the weather tool.
3When asked to calculate, use the calculator tool.
4Always explain your reasoning.

1// Tool node configuration
2URL: https://api.weatherapi.com/v1/current.json
3Method: GET
4Query Parameters:
5  - key: {{ $env.WEATHER_API_KEY }}
6  - q: {{ $json.location }}

1Input: "What's the weather in Hanoi and calculate 15% tip on $50"
2 
3Agent reasoning:
41. User wants weather → Use weather tool for Hanoi
52. User wants calculation → Use calculator for 50 * 0.15
63. Combine results
7 
8Output: "The weather in Hanoi is 28°C with partly cloudy skies. 
9        A 15% tip on $50 would be $7.50."

1// Memory node configuration
2Type: Window Buffer Memory
3Window Size: 10  // Last 10 messages
4Session ID: {{ $json.sessionId }}

1Type: Summary Memory
2Model: GPT-3.5 (for summarization)
3Max Tokens: 500

1Type: Vector Store Memory
2Embedding Model: OpenAI Embeddings
3Vector Store: Pinecone/Supabase
4Top K: 5  // Retrieve 5 most relevant

1// Custom tool với Code node
2// Description: "Search products by name and category"
3
4const input = $input.first().json;
5const { productName, category } = input;
6
7// Simulate database search
8const products = [
9  { id: 1, name: "iPhone 15", category: "phones", price: 999 },
10  { id: 2, name: "MacBook Pro", category: "laptops", price: 1999 },
11  { id: 3, name: "AirPods Pro", category: "audio", price: 249 }
12];
13
14let results = products;
15
16if (productName) {
17  results = results.filter(p => 
18    p.name.toLowerCase().includes(productName.toLowerCase())
19  );
20}
21
22if (category) {
23  results = results.filter(p => p.category === category);
24}
25
26return [{ json: { products: results } }];

1// PostgreSQL query tool
2// Input: { query: "user question about data" }
3
4const userQuestion = $input.first().json.query;
5
6// Agent constructs SQL based on question
7// Tool executes and returns results
8
9// Example: "How many orders last month?"
10// → SELECT COUNT(*) FROM orders WHERE created_at > NOW() - INTERVAL '1 month'

1You are a research assistant. When given a topic:
2 
31. First, search the web for recent news/articles
42. Then, check Wikipedia for background information
53. Finally, synthesize findings into a clear report
6 
7Format your report as:
8## Topic Overview
9[Background from Wikipedia]
10 
11## Recent Developments
12[From web search]
13 
14## Key Takeaways
15[Your synthesis]

1// Trong tool node
2try {
3  const result = await callExternalAPI();
4  return [{ json: { success: true, data: result } }];
5} catch (error) {
6  return [{ json: { 
7    success: false, 
8    error: error.message,
9    suggestion: "Try with different parameters"
10  }}];
11}

1System Prompt addition:
2 
3If a tool fails, explain the issue to the user and suggest alternatives.
4Never make up information - if you can't find data, say so.

1// System Prompt
2const systemPrompt = `
3You are a customer support agent for TechStore.
4
5Available tools:
6- order_lookup: Find order by ID or email
7- product_info: Get product details
8- refund_request: Submit refund request
9- escalate: Transfer to human agent
10
11Guidelines:
12- Always greet customers warmly
13- Look up order information before answering questions
14- For complex issues, offer to escalate
15- Never share other customers' information
16`;