This is the source code for my blog post Building AI Multi-agent Workflows.

It's a simple implementation of a multi-agent workflow to analyze stock market data and make investment recommendations. It uses the AutoGen framework. See my blog post above for more details.

The AI agents in the workflow use an Azure OpenAI model client. This means that you need to have an Azure subscription and an Azure OpenAI resource with a model deployed (in Azure AI Foundry). In this case, we're using gpt-4o.

If you want to use a different model, modify the get_azure_openai_chat_completion_client() method on line 185 in agents_workflow.py to reflect this.

Important: Wether you use a different model or not, make sure the parameters to AzureOpenAIChatCompletionClient() on line 191 in agents_workflow.py match your Azure OpenAI's deployment information.

Once you have your AI model deployment, set an environment variable as:

MY_AZURE_OPENAI_ENDPOINT="[YOUR_AZURE_OPENAI_SERVICE_ENDPOINT]"

Replacing the "[YOUR_AZURE_OPENAI_SERVICE_ENDPOINT]" placeholder with your own Azure OpenAI Service endpoint (you can find this information on your Azure AI Foundry dashboard).

Additionally, our model client uses Azure Active Directory (AAD) authentication. You need to assign the Cognitive Services OpenAI User role to your Azure user.

Once all of the above is configured, make sure you're signed in to your Azure account from your local environment. To sign-in, use the Azure cli on your terminal. Sign-in using the tenant ID for your user's Microsoft Entra ID as follows (search for "Microsoft Entra ID" in the Azure portal and copy the tenant ID).

az login --tenant [YOUR_TENANT_ID]

Replace the [YOUR_TENANT_ID] placeholder with your own tenant ID.

However, if you prefer to use other AI model clients—e.g. Gemini, OpenAI, a model hosted on Anthropic, or from Ollama—you'll need to modify the client creation method in the source code.

Refer to AutoGen's docuementation here to implement the new client creation method. Then assign the client to the model_client variable on line 202 in agents_workflow.py.

Note: in this case you don't need to create the environment variable MY_AZURE_OPENAI_ENDPOINT as mentioned above.

The data retrieval tool uses the sec-api python library, which requires an API key. You can obtain a free API key from here. Once you have your key, create an environment variable as:

MY_SEC_API_KEY="[YOUR_API_KEY]"

Replacing the placeholder with the value of your API key.

Finally, update the value of the constants on lines 7 and 8 in agents_tools.py with your name and email address. This is needed in order to query the SEC's data API.

7 HTTP_USER_AGENT_NAME = "[YOUR NAME]"
8 HTTP_USER_AGENT_EMAIL = "[YOUR EMAIL]"

To run the code, clone this repository to your machine. Then create a python virtual environment as follows.

Important: AutoGen requires python version 3.10 or higher.

1. On the terminal, navigate to the repository's directory. Then execute:

   python -m venv <venv_name>

   where <venv_name> is the name you want to give the virtual environment

2. To activate the environment:

   On MacOS/Linx execute:

   source <venv_name>/bin/activate

   On Windows execute:

   .\<venv_name>\Scripts\activate

3. With the virtual environment active, install the required libraries with pip:

   pip install -r ./requirements.txt

4. Now you can run the agents_workflow.py as:

     python3 agents_workflow.py

5. When you're done, deactivate the venv with:

   deactivate