Dmitry Kisselev
Initial deployment of customer support chatbot
efc63de
"""LLM agent with tool calling capabilities."""
import json
from typing import Dict, Any, List, Optional
from openai import OpenAI
from config import OPENAI_API_KEY, OPENAI_MODEL
from mcp_client import MCPClient
from auth import AuthHandler
class SupportAgent:
"""Customer support agent with MCP tool integration."""
def __init__(self, mcp_client: MCPClient, auth_handler: AuthHandler):
# Initialize OpenAI client
self.client = OpenAI(api_key=OPENAI_API_KEY)
self.model = OPENAI_MODEL
self.mcp_client = mcp_client
self.auth_handler = auth_handler
# Initialize MCP connection
self.mcp_client.initialize()
# Define available tools
self.tools = self._define_tools()
def _define_tools(self) -> List[Dict[str, Any]]:
"""Define tool schemas for OpenAI function calling."""
return [
{
"type": "function",
"function": {
"name": "list_products",
"description": "List products with optional filters by category or active status",
"parameters": {
"type": "object",
"properties": {
"category": {
"type": "string",
"description": "Filter by category (e.g., 'Computers', 'Monitors', 'Printers')"
},
"is_active": {
"type": "boolean",
"description": "Filter by active status"
}
}
}
}
},
{
"type": "function",
"function": {
"name": "get_product",
"description": "Get detailed product information by SKU",
"parameters": {
"type": "object",
"properties": {
"sku": {
"type": "string",
"description": "Product SKU (e.g., 'COM-0001', 'MON-0054')"
}
},
"required": ["sku"]
}
}
},
{
"type": "function",
"function": {
"name": "search_products",
"description": "Search products by name or description keyword",
"parameters": {
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "Search term (case-insensitive, partial match)"
}
},
"required": ["query"]
}
}
},
{
"type": "function",
"function": {
"name": "get_customer",
"description": "Get customer information by customer ID. Requires authentication.",
"parameters": {
"type": "object",
"properties": {
"customer_id": {
"type": "string",
"description": "Customer UUID"
}
},
"required": ["customer_id"]
}
}
},
{
"type": "function",
"function": {
"name": "list_orders",
"description": "List orders with optional filters. Requires authentication.",
"parameters": {
"type": "object",
"properties": {
"customer_id": {
"type": "string",
"description": "Filter by customer UUID"
},
"status": {
"type": "string",
"description": "Filter by status: draft, submitted, approved, fulfilled, cancelled"
}
}
}
}
},
{
"type": "function",
"function": {
"name": "get_order",
"description": "Get detailed order information including items. Requires authentication.",
"parameters": {
"type": "object",
"properties": {
"order_id": {
"type": "string",
"description": "Order UUID"
}
},
"required": ["order_id"]
}
}
},
{
"type": "function",
"function": {
"name": "create_order",
"description": "Create a new order with items. Requires authentication.",
"parameters": {
"type": "object",
"properties": {
"customer_id": {
"type": "string",
"description": "Customer UUID"
},
"items": {
"type": "array",
"description": "List of order items",
"items": {
"type": "object",
"properties": {
"sku": {"type": "string"},
"quantity": {"type": "integer"},
"unit_price": {"type": "string"},
"currency": {"type": "string", "default": "USD"}
},
"required": ["sku", "quantity", "unit_price"]
}
}
},
"required": ["customer_id", "items"]
}
}
}
]
def _requires_auth(self, tool_name: str) -> bool:
"""Check if tool requires authentication."""
auth_required_tools = ["get_customer", "list_orders", "get_order", "create_order"]
return tool_name in auth_required_tools
def _get_customer_id(self, session_id: str) -> Optional[str]:
"""Get customer_id from authenticated session."""
if not self.auth_handler.is_authenticated(session_id):
return None
return self.auth_handler.get_customer_id(session_id)
def process_message(self, session_id: str, user_message: str, conversation_history: List[Dict[str, str]]) -> str:
"""Process user message and return response."""
# Get authentication status
is_authenticated = self.auth_handler.is_authenticated(session_id)
customer_email = self.auth_handler.get_email(session_id) if is_authenticated else None
# Check if message is about authentication
if "email" in user_message.lower() and "pin" in user_message.lower():
# Try to extract email and PIN from message
# This is a simple approach - in production, use structured input
return "To authenticate, please provide your email and PIN in the format: 'email: your@email.com, pin: 1234'"
# Check if query might need authentication
order_keywords = ["order", "purchase", "buy", "my orders", "order history", "track order", "place order"]
needs_auth = any(keyword in user_message.lower() for keyword in order_keywords)
if needs_auth and not is_authenticated:
return "To access your orders, I need to verify your identity. Please provide your email and PIN in this format: 'email: your@email.com, pin: 1234'"
# Process with LLM (all API calls are automatically logged in OpenAI Platform under Logs → Completions)
response_text = self._process_with_llm(session_id, user_message, conversation_history, is_authenticated, customer_email)
return response_text
def _process_with_llm(self, session_id: str, user_message: str, conversation_history: List[Dict[str, str]], is_authenticated: bool, customer_email: Optional[str]) -> str:
"""Internal method to process message with LLM."""
# Build system message with authentication status
auth_status = "authenticated" if is_authenticated else "not authenticated"
system_content = """You are a helpful customer support agent for a computer products company.
You can help customers with:
- Product inquiries (browsing, searching, getting details) - no authentication needed
- Order management (viewing orders, order status, placing orders) - requires authentication
Current session status: """ + auth_status
if customer_email:
system_content += f"\nAuthenticated customer: {customer_email}"
system_content += """
IMPORTANT INSTRUCTIONS:
- When a customer asks to see/list/show their orders, use the list_orders tool directly
- When a customer asks about a specific order, use the get_order tool
- The customer_id is already set for authenticated sessions - you don't need to provide it
- Be friendly, professional, and helpful. Provide clear, concise answers."""
messages = [
{
"role": "system",
"content": system_content
}
]
# Add conversation history
messages.extend(conversation_history)
# Add current user message
messages.append({"role": "user", "content": user_message})
try:
# Call OpenAI with tool calling
# Note: For standard OpenAI Python SDK, API calls appear in Logs -> Completions
# The Traces tab is for OpenAI Agents SDK (JavaScript/TypeScript)
response = self.client.chat.completions.create(
model=self.model,
messages=messages,
tools=self.tools,
tool_choice="auto"
)
message = response.choices[0].message
# Handle tool calls
if message.tool_calls:
tool_results = []
for tool_call in message.tool_calls:
tool_name = tool_call.function.name
tool_args = json.loads(tool_call.function.arguments)
# Check authentication for order-related tools
if self._requires_auth(tool_name) and not self.auth_handler.is_authenticated(session_id):
tool_results.append({
"role": "tool",
"tool_call_id": tool_call.id,
"name": tool_name,
"content": "Authentication required. Please provide your email and PIN."
})
continue
# Inject customer_id for order-related tools
if self._requires_auth(tool_name):
customer_id = self._get_customer_id(session_id)
if customer_id:
# ALWAYS replace customer_id with the authenticated UUID
# Don't trust what the LLM provides - it may provide email instead
if tool_name in ["list_orders", "get_customer", "create_order"]:
tool_args["customer_id"] = customer_id
else:
# If customer_id is not available, don't call the tool
# This prevents using email as customer_id
tool_results.append({
"role": "tool",
"tool_call_id": tool_call.id,
"name": tool_name,
"content": "Error: Customer ID not found. Please re-authenticate."
})
continue
# Call MCP tool
try:
result = self.mcp_client.call_tool(tool_name, tool_args)
# Extract text content from result
if "content" in result and len(result["content"]) > 0:
content = result["content"][0].get("text", str(result))
elif "structuredContent" in result:
content = result["structuredContent"].get("result", str(result))
else:
content = str(result)
tool_results.append({
"role": "tool",
"tool_call_id": tool_call.id,
"name": tool_name,
"content": content
})
except Exception as e:
tool_results.append({
"role": "tool",
"tool_call_id": tool_call.id,
"name": tool_name,
"content": f"Error: {str(e)}"
})
# Get final response with tool results
messages.append(message)
messages.extend(tool_results)
# Final response - automatically traced
final_response = self.client.chat.completions.create(
model=self.model,
messages=messages
)
return final_response.choices[0].message.content
else:
return message.content
except Exception as e:
return f"I apologize, but I encountered an error: {str(e)}. Please try again."