Building my First Agent

This tutorial will guide you through the process of creating a simple agent and testing it using pytest.

1 Define the Agent's Job

First, let's create a new file my_agent.py. We'll define a class for our agent that inherits from crewmaster.agent.AgentBase. We need to provide a job_description to define its purpose.

# my_agent.py

from crewmaster.agent import AgentBase
from langchain_openai import ChatOpenAI
from pydantic import BaseModel

# A simple input model for our agent
class SimpleInput(BaseModel):
    query: str

class SimpleAgent(AgentBase):
    job_description = "You are a simple agent that responds to greetings."

2 Create and Test the Agent

Next, let's create a pytest file test_my_agent.py to test our agent's behavior.

# tests/test_my_agent.py

import pytest
from langchain_openai import ChatOpenAI
from crewmaster.core.runnables import run_in_thread
from my_agent import SimpleAgent, SimpleInput

@pytest.fixture
def llm():
    return ChatOpenAI(model_name="gpt-3.5-turbo")

def test_simple_agent_greeting(llm):
    agent = SimpleAgent(llm_service=llm)

    # We use a simple input model for our query
    input_data = SimpleInput(query="Hello, how are you?")

    # The agent's invoke method runs in a separate thread
    result = run_in_thread(agent.invoke, input_data)

    # We check if the agent's response contains a friendly greeting
    assert "Hello" in result.message.response or "Hi" in result.message.response

def test_simple_agent_unknown_query(llm):
    agent = SimpleAgent(llm_service=llm)

    input_data = SimpleInput(query="What is the capital of France?")

    result = run_in_thread(agent.invoke, input_data)

    # We expect the agent to defer or state it cannot answer
    assert "I cannot answer that question" in result.message.response

3 Run the Test

Run pytest from your terminal to execute the test.

pytest tests/test_my_agent.py

If the tests pass, you have successfully created and tested a basic agent!