Create No Code Powerful AI Agents – Azure AI Foundry

-

Introduction

An AI agent is a smart program that can think, make decisions, and do tasks. Sometimes it works alone, and sometimes it works with people or other agents. The main difference between an agent and a regular assistant is that agents can do things on their own. They don’t just help—you can give them a goal, and they’ll try to reach it.

Every AI agent has three main parts:
  • Model: This is the brain that helps the agent understand language and think.
  • Instructions: These tell the agent what it should do and what rules to follow.
  • Tools: These help the agent find information or take action.
Agents can take input like a message or a prompt and respond with answers or actions. 
For example, they might look something up or start a process based on what you asked.

Azure AI Foundry is a platform that brings all these things together; so you can build, train, and manage AI agents easily.

References

Usage

Firstly, we create a project in Azure AI Foundry.


Click on Next and give a name to your project.


Wait till the setup finishes.


Once the project creation finishes we are greeted with this screen.


Click on Agents tab and click on Next to choose the model.


I'm currently using GPT-4o Mini. It also includes descriptions for all the available models.


Then we configure the deployment details.
There are multiple deployment types available such as -
Global Deployments
  1. Global deployments [GlobalStandard] use Azure’s worldwide network to automatically send requests to the most available data center.
    This provides high usage limits and removes the need for manual load balancing.
    For apps with heavy, steady traffic, response times may vary.
    To reduce delays and ensure consistent performance, consider using provisioned throughput.
  2. Global Provisioned [GlobalProvisionedManaged] deployments also use Azure’s global network but include reserved processing power.
    They're designed for high and predictable workloads, offering better performance and reliability at scale.
  3. Global Batch [GlobalBatch] is made for processing large amounts of data efficiently and at a lower cost—around 50% cheaper than global standard.
    Instead of sending requests one by one, you send them all at once in a file. The system processes them asynchronously within 24 hours.
    It also uses a separate quota, so it doesn’t affect your live traffic.
Data Zone Standard Deployments
  1. Data Zone Standard [DataZoneStandard] deployments send requests to the best data center within a Microsoft-defined data zone, using Azure's global network.
    They offer higher default quotas than region-based setups. However, if your traffic is steady and high, response times may vary.
    For more stable performance, go with provisioned deployments.
  2. Data Zone Provisioned [DataZoneProvisionedManaged] deployments also operate within a Microsoft-defined data zone but with reserved model capacity.
    This ensures more consistent and reliable performance, especially for high-volume workloads.
  3. Data Zone Batch [DataZoneBatch] deployments work just like Global Batch, but all processing happens inside a specific Microsoft-defined data zone.
    It’s ideal for large-scale tasks where data residency within that zone is important.
Standard deployments [Standard] follow a pay-per-use model perfect for getting started quickly.
They're best for low to medium usage with occasional traffic spikes. 
However, for high and steady loads, performance may vary.

Provisioned deployments [ProvisionedManaged] let you pre-allocate the amount of processing power you need.
This is measured using Provisioned Throughput Units (PTUs). 
Each model and version requires a different number of PTUs and offers different performance levels.
Provisioned deployments ensure predictable and stable performance for large or mission-critical workloads.

This is how the deployment details look for in Global Standard.


I'll be choosing Standard deployment for our use case.


Click on deploy and wait for a few seconds.


Once the deployment is completed, you can give your agent a name and some instructions for their behavior. You should specify the tone, end goal, verbosity, etc as well.


You can also specify the Temperature and Top P values which are both a control on the randomness or creativeness of the model.
Temperature controls how bold or cautious the model is.
Lower temperature = Safer, more predictable answers. (Factual Q&A, Code Summarization)
Higher temperature = More creative or surprising answers. (Poetry/Creative writing)

Top P (Nucleus Sampling) controls how wide the model’s word choices are.
Lower Top P = Only picks from the most likely words. (Legal or financial writing)
Higher Top P = Includes less likely, more diverse words. (Brainstorming names)


Next, I'll add a knowledge base to my bot.


For this example, I'll just upload a single file.
However, you have the option to add an sharepoint folder or files, connect it to Bing Search, MS Fabric, Azure AI search, etc as required.


A Vector store in Azure AI Foundry helps your AI agent retrieve relevant information based on meaning rather than just keywords.
It works by breaking your content (like a PDF) into smaller parts, converting them into numerical representations (embeddings), and storing them.
When a user asks a question, the AI finds the most semantically similar parts from the vector store and uses them to generate accurate, context-aware responses.


Once you select the file, click on Upload and save.


At this point, you can start to interact with your model.
To "play around" with your model, click on the "Try in Playground" button.


And here, we can see the output based on our provided knowledge base.


One more example, just because it is kind of fun.


Every input that you provide to the agent is called as a "message".
Everytime the agent is invoked for processing the provided input is called a "run".
Every interaction session with the agent is called a "thread".

We can see all the open threads in the threads section.

Conclusion

Azure AI Foundry makes it easy to build and use AI agents without writing any code. 
You can choose models, set how they behave, and connect your own data all through a simple interface. Whether you're testing ideas, automating tasks, or building custom bots, Foundry gives you the tools to do it.
If you’re curious about AI or want to try building your own agent, Foundry is a great place to begin.

Comments

Post a Comment

Popular posts from this blog

Event-Driven Architecture: How to Create Custom Business Events in Business Central

-
Image
Introduction Events are notifications that signal something has happened in the system. A record getting created, updated or deleted, a sales order getting posted, etc are all examples of events. An event driven architecture provides asynchrony and decoupled communication between different components of the system supporting flexibility, scalability and modularity. In Business Central, we have two main types of events - Business and Integration. Integration Events are related to things within Business Central. They are use for extending the functionality of the Base Application without modifying it directly. Business Events are related to things outside of Business Central. They are used for notifying external systems (e.g., via Power Automate or Logic Apps) regarding any changes that happen within Business Central. Microsoft allows us to create custom Integration and Business Events based on our requirements. In this blog, we'll see how to create a custom Business Event and...
Read more

Visualizing Data: How to Add Power BI Reports to Business Central

-
Image
Introduction Power BI is a great tool for turning data into clear, interactive reports and the best part?  It works smoothly with Business Central, right out of the box.  You just need to set it up, and you can start viewing powerful reports right inside within Business Central dashboard.  Microsoft provides several ready-made reports, grouped into different apps, so you can pick and install only what you need.  Once set up, these reports help you track key business insights without switching between systems.  In this blog, we’ll walk you through how to set up and use Power BI reports in Business Central to make smarter decisions. References Introduction to Business Central and Power BI Install Power BI apps for Business Central Configuration Open your Business Central and search for "Assisted Setup". Click on "Connect to Power BI" Once the set up page opens, click on Next. Fiscal : A 12-month calendar that begins in any month and ends 12 months after. Standard ...
Read more

Know Your Users: Automated Usage Tracking for Better System Adoption

-
Image
Introduction After a new Business Central system goes live, it’s important to know if users are actually using it the way they should (Go-Live is just the beginning). Tracking usage statistics helps us understand how well the system is being adopted and where improvements might be needed. To make this easier, I’ve created a small utility that automatically collects user activity data and builds a report from it. This saves time and gives a clear picture of how the system is being used to the management. Here’s what it does: There’s a setup page where you can choose how stats are collected and skip certain users (like consultants or support team members who create test data). A configuration page lets you pick which tables to track. You can also set how often stats should be gathered using a frequency field; it even sets up the background job for you. A report which summarises the generated data in a neat matrix. With this tool, keeping an eye on system adoption becomes simple and autom...
Read more