Getting started with AutoGen framework for building AI agents and applications

1
2
3
4
5

$ mkdir -p autogen-101 && cd autogen-101
$ python -m venv .venv
$ source .venv\bin\activate
$ pip install -U "autogen-agentchat"
$ pip install "autogen-ext[openai,azure]"

1
2
3
4
5

from dotenv import load_dotenv
import os
import asyncio

load_dotenv()

1
2

from autogen_ext.models.openai import AzureOpenAIChatCompletionClient
from autogen_core.models import UserMessage, AssistantMessage, SystemMessage

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19

async def main():
    aopenai_client = AzureOpenAIChatCompletionClient(
        model=os.getenv("AZURE_OPENAI_CHAT_DEPLOYMENT_NAME"),
        api_key=os.getenv("AZURE_OPENAI_API_KEY"),
        api_version=os.getenv("AZURE_OPENAI_API_VERSION"),
        endpoint=os.getenv("AZURE_OPENAI_ENDPOINT"),
    )

    result = await aopenai_client.create([
        UserMessage(
            content="What is AutoGen framework from Microsoft Research?",
            source="user",
        ),
    ])

    print(result.content)
    
if __name__ == "__main__":
    asyncio.run(main())

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32

$ python 01-model-client.py
AutoGen is an advanced framework developed by Microsoft Research that facilitates the creation of complex agent-based workflows centered around Large Language Models (LLMs). Released as an open-source Python library, AutoGen provides developers with tools to streamline communication and collaboration between LLM-powered agents while enabling dynamic automation and interaction with external APIs or services.

### Key Features of AutoGen:
1. **Agent Collaboration and Communication**: AutoGen allows for the orchestration of intelligent agents (e.g., user agents or assistant agents) that can dynamically interact through dialogues and tasks. These agents are designed to perform complex reasoning, problem-solving, and coordination.

2. **Extensibility and Customization**: AutoGen provides a simple yet powerful interface where developers can define custom agents with specific behaviors, abilities, and APIs. This makes it highly adaptable for diverse use cases.

3. **Streamlined Workflow Automation**: Using AutoGen, developers can build sophisticated workflows where multiple agents collaborate to achieve tasks like information retrieval, code generation, decision-making, content creation, and more.

4. **Ease of Use**: While enabling complex logic and inter-agent collaboration, AutoGen emphasizes simplicity, reducing the time and effort required for developers to design, implement, and manage multi-agent systems.

5. **Hybrid Intelligence**: The framework bridges LLMs and external tools, allowing agents to integrate well with domain-specific APIs, databases, and other services for enhanced operational capabilities.

### Example Use Cases:
- **Code Development**: Agents can collaborate to design, write, debug, and review code based on user requirements.
- **Business Process Automation**: AutoGen can automate workflows in customer support, data analysis, and report generation by leveraging human-like decision-making agents.
- **Creative Content Generation**: Agents can coordinate for tasks requiring writing stories, generating marketing material, or brainstorming ideas collaboratively.

### Components of AutoGen:
- **UserAgent**: Simulates a user interacting with the system.
- **AssistantAgent**: Represents an LLM-powered agent that actively solves problems or performs delegated tasks.
- **Custom Agents**: Developers can design unique agents tailored to specific roles or actions, integrating APIs, static data, or dynamic processes.
- **Robust Communication Framework**: Facilitates seamless back-and-forth interaction between agents, with dialogue and task-driven workflows.

### Availability:
Microsoft Research released AutoGen on GitHub as an open-source project. Developers can freely access the library, explore examples, and adapt its features to their needs.

### Significance:
AutoGen highlights the growing trend of transitioning from single-agent LLM applications to multi-agent systems capable of collaborative intelligence. Such frameworks push the boundaries of what LLMs can achieve, offering opportunities for groundbreaking innovation in AI-driven automation and human-computer interaction.

You can find the library and documentation for AutoGen [here on GitHub](https://github.com/microsoft/autogen).

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

from dotenv import load_dotenv
import os
import asyncio

from autogen_ext.models.ollama import OllamaChatCompletionClient
from autogen_core.models import UserMessage

load_dotenv()

async def main():
    ollama_client = OllamaChatCompletionClient(
        model="llama3.2:3b"
    )

    result = await ollama_client.create([
        UserMessage(
            content="What is AutoGen framework from Microsoft Research?",
            source="user",
        ),
    ])

    print(result.content)

if __name__ == "__main__":
    asyncio.run(main())

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21

$ python .\01-model-client-ollama.py
AutoGen is a tool developed by Microsoft Research that enables the automatic generation of test cases for software applications. It is designed to help improve the quality and coverage of unit tests, integration tests, and UI tests.

AutoGen uses a combination of machine learning algorithms and natural language processing (NLP) techniques to analyze source code, identify relevant features, and generate test cases that cover those features. The goal is to reduce the manual effort required to write test cases, making it easier for developers to ensure their application's functionality is thoroughly tested.

Here are some key features of AutoGen:

1. **Source code analysis**: AutoGen analyzes source code to identify relevant functions, methods, and classes.
2. **Feature extraction**: It extracts relevant features from the analyzed source code, such as input parameters, return types, and data structures.
3. **Test case generation**: Based on the extracted features, AutoGen generates test cases that cover those features.
4. **Test case optimization**: It optimizes generated test cases to minimize duplication and ensure maximum coverage.

AutoGen can be used in various testing scenarios, including:

1. Unit testing: Generating test cases for individual functions or methods.
2. Integration testing: Testing interactions between components or modules.
3. UI testing: Creating test cases for user interface elements and behaviors.

By leveraging machine learning and NLP techniques, AutoGen aims to make the testing process more efficient and effective, allowing developers to focus on other aspects of software development.

Would you like to know more about how AutoGen works or its applications in specific industries?

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73

from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.ui import Console
from autogen_ext.models.openai import AzureOpenAIChatCompletionClient

import asyncio
from dotenv import load_dotenv
import os
import json
import requests
from datetime import datetime

from typing import Any

load_dotenv()

api_key = os.getenv("AZURE_OPENAI_API_KEY")
model = os.getenv("AZURE_OPENAI_CHAT_DEPLOYMENT_NAME")
endpoint = os.getenv("AZURE_OPENAI_ENDPOINT")
api_version = os.getenv("AZURE_OPENAI_API_VERSION")
aoi_client = AzureOpenAIChatCompletionClient(
    model=model,
    api_key=api_key,
    endpoint=endpoint,
    api_version=api_version,
)

def get_weather(city: str, date: datetime = None) -> dict[str, Any]:
    """
    Get the weather at a given location on a given date or current weather.

    Args:
        city: The city name, e.g. Bengaluru.
        date: Date on which the weather at the given location should be determined. This defaults to the current weather when a date is not specified.

    Returns:
        JSON string with the city name, date, and temperature.
    """
    api_key = os.getenv("VISUAL_CROSSING_API_KEY")
    if date is None:
        date = datetime.now().strftime("%Y-%m-%d")
    
    request_url = f"https://weather.visualcrossing.com/VisualCrossingWebServices/rest/services/timeline/{city}/{date}?unitGroup=metric&key={api_key}&contentType=json"
    response = requests.get(request_url)

    if response.status_code != 200:
        return json.dumps({
            "error": "Invalid city name or date"
        })
    else:
        respJson = response.json()
        return json.dumps({
            "city": city,
            "date": date,
            "temperature": respJson["days"][0]["temp"]
        })

agent = AssistantAgent(
    name="weather_agent",
    model_client=aoi_client,
    tools=[get_weather],
    system_message="You are a helpful assistant.",
    reflect_on_tool_use=True,
    model_client_stream=True,
)

async def main() -> None:
    await Console(
        agent.on_messages_stream(
            task="What is the weather in London?"
        )
    )

asyncio.run(main())

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

	"""
    Get the weather at a given location on a given date or current weather.

    Args:
        city: The city name, e.g. Bengaluru.
        date: Date on which the weather at the given location should be determined. This defaults to the current weather when a date is not specified.

    Returns:
        JSON string with the city name, date, and temperature.
    """
	return f"The weather in {city} on {date} is 30 degrees celsius and rainy."

1
2
3
4
5
6
7
8
9

$ python 02-assistant-agent.py
---------- user ----------
What is the weather in London?
---------- weather_agent ----------
[FunctionCall(id='call_Z4evOdxIYSuCsMyXEJXXyToR', arguments='{"city":"London"}', name='get_weather')]
---------- weather_agent ----------
[FunctionExecutionResult(content='{"city": "London", "date": "2025-03-23", "temperature": 11.0}', name='get_weather', call_id='call_Z4evOdxIYSuCsMyXEJXXyToR', is_error=False)]
---------- weather_agent ----------
The current temperature in London is 11°C. Let me know if you'd like more details about the weather!

1
2
3
4
5
6
7
8
9

$ python 02-assistant-agent.py
---------- user ----------
What is the weather in London?
---------- weather_agent ----------
[FunctionCall(id='call_1DXbEuxDnHKHKj6d6DhWmueo', arguments='{"city":"London"}', name='get_weather')]
---------- weather_agent ----------
[FunctionExecutionResult(content='{"city": "London", "date": "2025-03-23", "temperature": 11.0}', name='get_weather', call_id='call_1DXbEuxDnHKHKj6d6DhWmueo', is_error=False)]
---------- weather_agent ----------
{"city": "London", "date": "2025-03-23", "temperature": 11.0}