4. AI Agents

Agent Service Portal

Foundry's Agent Service provides a managed runtime for AI agents. You can create, configure, and monitor agents through the portal or SDK.

Python Agent Example

from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential

client = AIProjectClient(
    subscription_id="your-sub",
    resource_group="your-rg",
    project_name="my-project",
    credential=DefaultAzureCredential()
)

agent = client.agents.create_agent(
    model="gpt-4o",
    name="MyFirstAgent",
    instructions="You are a helpful assistant that answers questions about Azure."
)

response = client.agents.send_message(agent.id, "What is Azure AI Foundry?")
print(response.text)

Built-in Tools

Tool	Description
Web Searcher	Search the web for real-time information
Code Interpreter	Execute Python code in a sandboxed environment
OpenAPI Tool	Call any REST API via OpenAPI spec
MCP Server	Connect to MCP-compatible tool servers
Azure AI Search	Query vector indexes for RAG
Memory Store	Persist conversation history and user context

Multi-Tool Agent

from azure.ai.projects.models import ToolSet, CodeInterpreterTool, WebSearchTool

toolset = ToolSet()
toolset.add(CodeInterpreterTool())
toolset.add(WebSearchTool())

agent = client.agents.create_agent(
    model="gpt-4o",
    name="MultiToolAgent",
    instructions="Use tools to answer user questions.",
    toolset=toolset
)

Browser Automation

Foundry agents can control browser-based workflows for web scraping, form filling, and testing using Playwright-based automation tools.

Memory Store Concepts

Volatile — In-memory, session-scoped
File-based — Local persistence for development
Cosmos DB — Production-grade, distributed memory

✏️ Exercise: Create an agent with both Web Search and Code Interpreter tools. Ask it to "Search for the latest AI news and summarize the top 3 stories." Verify it uses both tools correctly.