Task 3 — Call your agent from a client app

Part of the Build and extend AI agents lab. New here? Start with Getting started.

Set up (start here): This task needs a Foundry project and the starter code. If you haven’t already, complete Getting started to create your project, clone the code, and set PROJECT_ENDPOINT in Python/.env.

This task drives a grounded agent. The quickest way to get one is to create it in code — from the Labfiles/A-build-and-extend-ai-agents folder, run:

python setup/bootstrap_agent.py

That creates and grounds tailwind-agent — including the Code Interpreter tool with the sales data already attached — and writes AGENT_NAME into your .env. Then verify you’re ready:

python setup/check_env.py --task 3

Already built the agent in Task 1? Use it instead of the script: open your tailwind-agent in the portal, add the Code interpreter tool with the sales data (step 1 below), and set AGENT_NAME=tailwind-agent in .env.

Goal: Interact with the grounded portal agent from a small web chat app instead of the playground — including charts the agent produces (from code interpreter), which render inline in the chat window.

Concept reinforced: consuming an agent programmatically with the Foundry SDK — loading an existing agent by name and driving it with the Responses API. A provided UI shell (tailwind_ui.py) turns your agent into a browser chat app, so you focus on the agent code, not the interface.

Set up:

If you ran python setup/bootstrap_agent.py above, your agent, its Code Interpreter tool, and AGENT_NAME are already configured — activate your virtual environment (.\labenv\Scripts\Activate.ps1) and skip to Try it first.

If you built the agent yourself in Task 1, finish wiring it up:

In the portal, open your tailwind-agent, add a Code interpreter tool, and upload a data file so there’s something to analyze. Download and attach:
```
 https://raw.githubusercontent.com/MicrosoftLearning/mslearn-ai-agents/main/Labfiles/A-build-and-extend-ai-agents/Python/weekly_sales.csv
```
Save the agent.
In the Labfiles/A-build-and-extend-ai-agents/Python folder, activate the virtual environment (.\labenv\Scripts\Activate.ps1). Then open .env and add AGENT_NAME=tailwind-agent alongside the PROJECT_ENDPOINT you already set. Save the file.

Try it first: The agent_with_functions.py file already contains a complete client that launches a web chat window. Before running it, predict: which SDK call loads your existing portal agent by name? How does the client tell the Responses API to use that agent? How does a respond() function turn one message into a reply the UI can show?

Show a solution

The provided agent_with_functions.py already implements the client and hands its respond() function to the shared run_chat_app() shell. The lines that matter are:

Load your portal agent by name (using AGENT_NAME from .env):

 agent = project_client.agents.get(agent_name=agent_name)

Route each request to that agent through the Responses API (inside respond()):

 response = openai_client.responses.create(
     conversation=conversation.id,
     extra_body={"agent_reference": {"name": agent.name, "type": "agent_reference"}},
     input="",
 )

Inline charts: helper functions detect image outputs and container_file_citation annotations, save them under agent_outputs/, and return them in an AgentReply so the UI renders them inline in the chat.
Launch the app: the file ends by starting the browser chat window:
```
 run_chat_app(respond, title="Tailwind Traders Assistant")
```

az login
python agent_with_functions.py

Your browser opens a chat window at http://localhost:7860. Ask for something that uses code interpreter:

Analyze the weekly sales data and create a chart of revenue over time.

The agent’s analysis appears in the chat and the chart is shown inline. Close the browser tab and press Ctrl+C in the terminal to stop the app.

Stretch: display the agent’s token usage after each response.

Next (optional): Task 4 — Add custom function tools