Instrument Your Agents

​

What is Agent Instrumentation?

Agent instrumentation is the process of enhancing your existing AI agent by integrating xpander’s backend capabilities. Rather than rebuilding your agent from scratch, you selectively replace key components with xpander’s services to gain advanced features like persistent memory, robust tool management, and multi-channel connectivity.

​

The Three Components to Instrument

When instrumenting your agent, focus on replacing these three essential components:

​

Benefits of Instrumentation

By instrumenting your agent with xpander, you gain:

1. Persistent Memory: Conversations are stored across sessions and accessible via API
2. Powerful Tools: Access to xpander’s tool library and execution framework
3. Multi-Interface Access: Connect to Slack, Discord, MCP, and other channels
4. Execution Monitoring: Track agent activity and performance
5. Agent Collaboration: Enable agent-to-agent delegation and communication

​

Set Up Your Environment

1. Install the xpander CLI:

   npm install -g xpander-cli

nvm install 22 && nvm use 22

2. Install required Python packages:

   # Create and activate virtual environment
   python3 -m venv .venv
   source .venv/bin/activate
   
   # Install dependencies
   pip install xpander-sdk xpander-utils openai

3. Authenticate with xpander:

   xpander login

4. Configure your environment: Create a .env file with:

   XPANDER_API_KEY=your_xpander_api_key
   XPANDER_ORG_ID=your_organization_id
   OPENAI_API_KEY=your_openai_api_key
   OPENAI_MODEL_ID=gpt-4.1

​

Create Your Agent Project

Run the agent creation wizard:

xpander agent new 
##  Or
xpander agent init

This command will guide you through creating a new agent and loading it into your current directory. The CLI generates a complete project structure:

agent_instructions.json  # Agent system prompt
Dockerfile               # Deployment configuration
requirements.txt         # Python dependencies
xpander_config.json      # Agent configuration
xpander_handler.py       # Event handler scaffold

​

Implementing Your Custom Agent

After setting up your project, you need to customize two key files:

​

1. Create Your Custom Agent Class

Create a file (e.g., myagent.py) that implements your agent’s core logic while leveraging xpander’s backend services:

from xpander_sdk import Agent, XpanderClient
from openai import OpenAI

class MyAgent:
    def __init__(self, backend: Agent) -> None:
        self.model = OpenAI()
        self.agent_backend = backend

    def chat(self, user_input: str, thread_id: str = None) -> str:
        # Let xpander handle thread/conversation management
        self.agent_backend.add_task(input=user_input, thread_id=thread_id)

        while not self.agent_backend.is_finished():
            # Call your LLM with your preferred parameters
            response = self.model.chat.completions.create(
                model="gpt-4.1",
                messages=self.agent_backend.messages,  # xpander provides context
                temperature=0.0,
                tools=self.agent_backend.get_tools(),  # xpander provides tools
                tool_choice=self.agent_backend.tool_choice
            )
            
            # Store conversation in xpander's persistent memory
            self.agent_backend.add_messages(response.model_dump())
            
            # Let xpander handle tool execution
            tool_calls = XpanderClient.extract_tool_calls(
                llm_response=response.model_dump()
            )
            tool_results = self.agent_backend.run_tools(tool_calls=tool_calls)
        
        # Get the final result from xpander
        final_result = self.agent_backend.retrieve_execution_result()
        return final_result.result

​

2. Customize the Event Handler

Edit the xpander_handler.py file generated by the CLI to integrate your custom agent:

..
from myagent import MyAgent  # Import your custom agent class
#...

@listener.register
def on_execution_request(execution_task: AgentExecution) -> AgentExecutionResult:
    # Get the xpander agent backend
    agent_backend = xpander.agents.get(agent_id=xpander_config.get("agent_id"))
    
    # Initialize the agent with the execution context
    agent_backend.init_task(execution=execution_task.model_dump()) 
    
    # Use your custom agent with xpander's backend
    my_agent = MyAgent(agent_backend)
    
    # Process the request and get the result
    result = my_agent.chat(user_input=execution_task.input.text)
    
    # Return the result to wherever the request originated
    return AgentExecutionResult(result=result)
#...

​

Testing Your Instrumented Agent

To test your agent:

1. Run locally:

   python xpander_handler.py

2. Access via web: Once running, your agent will be available at the URL displayed in the console.

3. Interact with your agent:

   • Send messages through the web interface
   • Watch as they’re processed locally with your custom logic
   • Observe the real-time logs in your terminal

​

Cloud Deployment

When you’re ready to deploy your agent to the cloud:

xpander deploy

This command packages your agent as a container and deploys it to the xpander cloud platform, making it accessible through all configured channels. You can monitor your deployed agent with:

xpander logs