Elasticsearch code examples for Microsoft Agent Framework

This repository contains some examples in Python and .NET for using the Microsoft Agent Framework with Elasticsearch.

This repository is part of the article Insert the title published in the Elasticsearch Labs website by Elastic.

Microsoft Agent Framework

Microsof Agetn Framework is a comprehensive multi-language framework for building, orchestrating, and deploying AI agents with support for both .NET and Python implementations. This framework provides everything from simple chat agents to complex multi-agent workflows with graph-based orchestration.

For more information you can read this Microsoft announcement.

Run Elasticsearch

To execute the examples reported in this repository you need to have an instance of Elasticsearch running. You can register for a free trial on Elastic Cloud or install a local instance of Elasticsearch on your computer.

To install locally, you need to execute this command in the terminal:

curl -fsSL https://elastic.co/start-local | sh

This will install Elasticsearch and Kibana on macOS, Linux and Windows using WSDL.

Use the sample data in Kibana

The examples reported in this repository use a sample data provided by Kibana. You need to import this sample data using the following procedure:

• After the login in Kibana, open the navigation menu on the left and select Integrations page in the navigation bar on the left (Figure 1).

• In the Integrations page, search for "sample" and click on Sample Data (Figure 2)

• Finally, in the Sample data page, click on Other sample data sets and click on Add data for the Sample eCommerce orders use case (Figure 3).

The ecommerce data will be stored in an index called kibana_sample_data_ecommerce containing about 4,675 orders.

Configure Azure AI

In the examples, We used Azure OpenAI. You need to configure a .env file containing the following environmental variables:

ELASTICSEARCH_ENDPOINT="The endpoint of your Elasticsearch deployment here"
ELASTICSEARCH_API_KEY="The API Key of your Elasticsearch deployment here"

AZURE_OPENAI_ENDPOINT="The Azure OpenAI endpoint here"
AZURE_OPENAI_RESPONSES_DEPLOYMENT_NAME="The Azure OpenAI deployment name here"
AZURE_OPENAI_API_KEY="The API Key of your Azure OpenAI deployment here"

You can generate the .env file copying if from .env-dev file.

If you installed Elasticsearch using start-local, you can read the endpoint and api key from the .env file of the folder installation.

For Azure OpenAI you can read the value settings in the Azure portal.

Python examples

To install the Python examples, you can create and activate a virtual environment (venv) in the python folder.

Use the following commands from the root folder of the repository:

cd python
python -m venv .venv
source .venv/bin/activate

After, you can install all the required packages as follows:

pip install -r requirements.txt

Now, you can execute the examples. For instance, you can run the simple_agent_tools.py with the following command:

python simple_agent_tools.py

This example build an agent that can interact with Elasticsearch using some tools. The tools are implemented in the python/elasticsearch_tools.py file.

.NET examples

To run the .NET examples, simply navigate to the directory of a specific example and start the program:

cd dotnet/{example name}
dotnet run

Please ensure that you set the necessary environment variables according to your setup.

License

This software is licensed under the Apache License 2.0