Azure AI Foundry service

Step 1: Setup Azure AI Foundry

Assuring you have known how to use Azure Portal and add azure service. I will skip the adding Azure OpenAI service.

Once you add Azure OpenAI service, open “Explore Foundry port” to open Foundry dashboard. the alternative uses https://ai.azure.com/

Recommend you switch to new Foundry. You will be asked either select a existed project or create a new project.

Creating new project is easy, simply follow the screen steps. you cannot miss it.

you can see that you are able to create agents, Explore playgrounds and Find modules and you recent done works.

1) Create Agents

Create Agents = Build your own AI assistant
创建一个“AI助手/智能体”

You use this when you want to:

• define a role (e.g. “Data Analyst Agent”)
• add instructions (system prompt) / 写规则
• connect tools (SQL, API, files) / 加工具（SQL / API / 文件）
• add knowledge (RAG) / 加知识库（RAG
• make it do tasks automatically

👉 Result: a custom AI agent / 造一个AI员工, 一个“能干活的AI”

2) Explore Playgrounds

Playgrounds = Testing area for models
砂箱系统， 模型试验室

You use it to:

• chat with models (GPT, DeepSeek, etc.) / 测试不同模型， GPT， DeepSeek .
• test prompts / 写 prompt 看效果
• try settings (temperature, tokens) / 调参数（temperature 等）
• compare responses / 做实验

👉 It is NOT production
👉 It is for experimenting
👉 It is a “Sandbox / practice room”

3) Find Models

Find Models = Choose AI model

You use it to:

• browse available models (GPT-4.x, GPT-5.x, DeepSeek, etc.) / 看有哪些可以用的模型（GPT、DeepSeek等）
• check capabilities / 对比能力
• compare cost/performance / 看价格，比性能
• decide which model to deploy

👉 It is the “model catalog” / 就是“挑AI大脑”

simply think as:

• Find Models → choose the brain / 选员工候选人
• Playgrounds → test the brain / 面试测试
• Create Agents → build a worker using the brain / 正式雇佣 + 分配工作

Step 2: Deploy model

From Project dashboard, click “Find models”, you will find many models over there to be selected. e.g. gpt-chat5.4, DeepSeek-V4-Flash, etc.

直白地说人话：安装一个model。

Choose a one you like, then click “Deploy”, “deploy” is done.

Step 3: Create Agent

From Project dashboard, click Create agents

follow steps to create a agent. It is straight forward. no any confusing. Fill in agent Name.

you create the agent. looks this:

Tool = giving the AI external capabilities.
给 AI 加“外部能力”. 没有tools， AI只能聊天-Chat， 有tools， AI才能做事。

From Agent UI, you can see:

What is “Create toolbox”

Create toolbox = create a container/group for tools

Inside toolbox you can later add:

• APIs
• Functions
• Search
• Database tools
• Custom tools

What is “Connect a tool”

Connect a tool = connect an actual usable tool/service/API

Examples:

• Bing Search
• Azure AI Search
• Function API
• REST API
• SQL
• OpenAPI service

First – Create toolbox

cleck “Create toolbox”

Second add tools to toolsBox

Then click Add to add tools into the toolBox

Let’s add a “Bing Search” as example.
cleck “Web search” –> Add tools

Add another Tool – Function / REST API, let Agent call external servicers.

{
  "openapi": "3.0.0",
  "info": {
    "title": "Users API",
    "version": "1.0.0"
  },
  "servers": [
    {
      "url": "https://jsonplaceholder.typicode.com"
    }
  ],
  "paths": {
    "/users": {
      "get": {
        "operationId": "getUsers",
        "summary": "Get list of users",
        "responses": {
          "200": {
            "description": "Successful response"
          }
        }
      }
    }
  }
}

now we have added 2 tools

Test the tools

Go to Agent playground: Agent → Chat / Playground / Test panel

Test 1: “REST API”

Typing “Use the REST API tool to get all users and show them.”

Test 2: Bing Search

Typing “Use Bing Search tool to find latest information about Azure AI Foundry.”

FORCE the Agent to call the tool

Assuming we have 100 REST API endpoints, each one will return different data, such as the user’s name or the company’s name, sale’s amount ……
When we add each API Endpoint to ToolBox, we have to give clearly, specifically descriptions. Agent will scan description, it will choose the most specific one to call,

In actual AI project, most case is using “Tag”,

e.g.
Tool Registry:
– name
– description
– schema
– tags

Tool: get_company_financials
Tags: finance, company, revenue, kpi

Tool: get_user_profile
Tags: user, identity, profile

Add below “instructure” On the “Agent UI” –> “Instructions”

“You have access to multiple tools.
Each tool has a description that defines its purpose.
Always:
– Read tool descriptions carefully
– Select the most relevant tool based on semantic meaning of the user request
– Do NOT rely on hardcoded routing rules
– If multiple tools are relevant, choose the most specific one”