Build Agent

Integrate PayLink payment tools into your AI agent

Now that you’ve tested your payment provider connection, you’re ready to integrate PayLink into your AI agent. This guide shows you how to build an agent using LangChain that can process payments for orders.

Install LangChain

Make sure you have LangChain installed:

pip

pip install langchain

Create Your Agent

1. Configure environment variables

   Create a .env file with the following environment variables:

   # OpenAI Configuration
   OPENAI_API_KEY="your_openai_api_key"
   
   # LangSmith Tracing
   LANGSMITH_API_KEY="your_langsmith_api_key"
   LANGCHAIN_TRACING_V2=true
   LANGCHAIN_PROJECT='payment-with-human-approval'
   
   # PayLink Configuration
   PAYLINK_API_KEY="your_paylink_api_key"
   PAYLINK_PROJECT="your_paylink_project_name"
   PAYLINK_TRACING="enabled"
   
   # Payment Provider
   PAYMENT_PROVIDER=["mpesa"]
   
   # M-Pesa Configuration
   MPESA_BASE_URL="https://sandbox.safaricom.co.ke"
   MPESA_CONSUMER_SECRET="your_mpesa_consumer_secret"
   MPESA_CONSUMER_KEY="your_mpesa_consumer_key"
   MPESA_PASSKEY="your_mpesa_passkey"
   MPESA_BUSINESS_SHORTCODE="your_business_shortcode"
   MPESA_CALLBACK_URL="https://yourdomain.com/mpesa-express-simulate/"
   

   Replace the placeholder values with your actual API keys and credentials.

2. Create the orders tool

   First, create a tool to retrieve orders from your database. This example uses a simple in-memory database:

   from langchain.tools import tool
   from typing import Literal
   
   TOOL_DESCRIPTION = """
   This tool returns orders from the database
   
   Args:
   - payment_status: The payment status of the orders. Can be "paid", "pending", or "failed". If not provided, all orders are returned.
   
   Returns:
   - A list of orders with the following fields:
       - name: The name of the order
       - price: The price of the order
       - quantity: The quantity of the order
       - payment_status: The payment status of the order
   """
   
   @tool("get_orders", description=TOOL_DESCRIPTION)
   def get_orders(
       payment_status: Literal["paid", "pending", "failed"] | None = None,
   ) -> list[dict]:
       """Return orders in the cart, optionally filtered by payment status."""
   
       orders = [
           {
               "name": "Item 1",
               "price": 100,
               "currency": "Ksh",
               "quantity": 1,
               "payment_status": "paid",
           },
           {
               "name": "Item 2",
               "price": 200,
               "currency": "Ksh",
               "quantity": 2,
               "payment_status": "pending",
           },
       ]
   
       if payment_status is None:
           return orders
   
       normalized_status = payment_status.lower()
       return [
           item for item in orders if item["payment_status"].lower() == normalized_status
       ]
   
3. Initialize PayLink and create the agent

   Set up PayLink tools and create your agent:

   from langchain.agents import create_agent
   from langchain.chat_models import init_chat_model
   from paylink.integrations.langchain_tools import PayLinkTools
   from src.tools.get_orders import get_orders
   
   paylink_client = PayLinkTools()
   payment_tools = paylink_client.list_tools()
   
   tools = [get_orders] + payment_tools
   
   agent = create_agent(
       model=init_chat_model(model="gpt-4o-mini"),
       tools=tools,
   )
   
4. Running the Agent

   Start the LangGraph development server:

   langgraph dev
   

   This will start the agent and make it available at http://localhost:2024.

5. Interacting with the Agent

   Once the server is running, you can interact with your agent through the LangGraph Studio interface or via API calls.

6. Example conversation flow

   Here’s an example of how a conversation with your agent might flow:

   1. Agent: “I can help you with orders and payments. What would you like to do?”

   2. User: “Show me pending orders”

   3. Agent: Uses get_orders tool with payment_status="pending"

   4. Agent: “I found 2 pending orders. Would you like me to process payment for any of them?”

   5. User: “Yes, process payment for Item 2”

   6. Agent: Prepares payment request and pauses for approval

   7. System: “Approval Required: Review the payment request for Item 2 (400 Ksh)”

   8. User: Reviews and approves

   9. Agent: Executes payment via PayLink

   

View Execution Traces

You can view detailed execution traces of your agent’s interactions in LangSmith. This allows you to inspect tool calls, model responses, and the complete execution flow.

This is an example trace:

View example trace →

How It Works

The agent is configured with:

• Payment Tools: PayLink tools (like stk_push) are automatically available to the agent
• Order Management: The get_orders tool allows the agent to retrieve orders that need payment

When a user asks the agent to pay for pending orders, the agent will:

1. Retrieve pending orders using get_orders
2. Calculate the total amount
3. Initiate payment using the stk_push tool
4. The user receives a payment prompt on their phone to complete the transaction

Full Code Example

For a complete, working example of an agent with PayLink payment integration, see the agent-with-payment repository on GitHub.

Next Steps

Once your agent can handle payments, learn how to implement Human in the Loop workflows for payment approvals.