Nat Taylor — Blog, AI, Product Management & Tinkering

Test Drive: Swarm

Published on Oct 14, 2024.

Today I’m test driving swarm, an “Educational framework exploring ergonomic, lightweight multi-agent orchestration. Managed by OpenAI Solution team.” My task will be to get the agents to take a customer problem, define requirements, and write some code. The process is simple:

1. Define your agent’s instructions
2. Define any tools
3. Run the swam

Adding the step for the PM to define requirements does seem to help the SWE agent. I like that it prints everything so you can see the tool calls that are being made.

from swarm import Swarm, Agent
from rich import print

client = Swarm()

def ask_for_code(requirements):
    return swe

pm = Agent(
    name="Product Manager",
    instructions = "Based on the problem statment, you produce simple requirements for the software engineer.  You ask the software engineer for code with the requirements you determine.",
    functions=[ask_for_code],
)

swe = Agent(
    name="Software Engineer",
    instructions="You are a python Flask expert who implements solutions in python Flask.  You produce code.",
)

response = client.run(
    agent=pm,
    messages=[{"role": "user", "content": "It takes too long to get a weather forecast, so I need you to implement a script that calls the NWS weather API"}],
    model_override='gpt-4o-mini',
)Code language: Python (python)

Results

Certainly! Below is a Python Flask application that calls the NWS (National Weather Service) weather API to get the current weather forecast based on user-provided latitude and longitude. The application handles errors gracefully and returns the current temperature and weather conditions.

from flask import Flask, request, jsonify
import requests

app = Flask(__name__)

NWS_API_BASE_URL = "https://api.weather.gov/points"

def get_weather_forecast(latitude, longitude):
    # Getting the forecast URL from NWS API using latitude and longitude
    forecast_url = f"{NWS_API_BASE_URL}/{latitude},{longitude}"
    
    try:
        response = requests.get(forecast_url)
        response.raise_for_status()  # Raise an error for bad responses
        forecast_data = response.json()
        
        # Extract the forecast URL from the response
        forecast_api_url = forecast_data['properties']['forecast']
        
        # Fetching the actual forecast
        forecast_response = requests.get(forecast_api_url)
        forecast_response.raise_for_status()
        forecast_details = forecast_response.json()
        
        # Extracting current conditions
        current_conditions = forecast_details['properties']['currentObservations'][0]
        temperature = current_conditions['temperature']
        weather = current_conditions['shortForecast']
        
        return {
            'temperature': temperature,
            'weather': weather
        }
    except requests.exceptions.HTTPError as http_err:
        return {'error': f'HTTP error occurred: {http_err}'}, 400
    except Exception as err:
        return {'error': f'An error occurred: {err}'}, 400

@app.route('/weather', methods=['GET'])
def weather():
    # Getting latitude and longitude from query parameters
    latitude = request.args.get('lat')
    longitude = request.args.get('lon')
    
    if not latitude or not longitude:
        return jsonify({'error': 'Please provide latitude and longitude as query parameters.'}), 400
    
    weather_data = get_weather_forecast(latitude, longitude)
    
    return jsonify(weather_data)

if __name__ == '__main__':
    app.run(debug=True)Code language: Python (python)

Instructions to Run:

1. Install Flask and Requests: Make sure you have
   the Flask and requests libraries installed.
   You can install them using pip:

   
   

   pip install Flask requests

2. Save the code: Save the above code in a file
   named app.py.

3. Run the Flask app:

   
   

   python app.py

4. Accessing the API: You can access the weather
   forecast by going to the following endpoint in your web browser or using
   a tool like Postman:

   
   

   http://127.0.0.1:5000/weather?lat=38.4247341&lon=-86.9624086

   
   

   Make sure to replace lat and lon with the
   desired latitude and longitude values.

This application will return the current temperature and weather conditions for the specified location.