Skip to main content
CanoeUX home

SharePoint Web Parts Essentials

CNX Azure AI Setup

Connect Azure AI once for every CNX web part: deploy the SharePoint AI Gateway, connect it, check it, and add or change AI models from one SharePoint page.

Overview

CNX Azure AI Setup is the one place where a SharePoint administrator connects Azure AI to the CNX web parts. Set it up once, and every CNX web part that uses AI (such as CNX Icon Studio, CNX Image Studio, CNX SharePoint Control+, CNX User Diagnostics and the message editor shared by several web parts) can use it, with no setup of its own.

It guides you through three steps: deploy the SharePoint AI Gateway to your own Azure subscription, connect it to all the web parts, and check that it works. Afterwards, come back to the same page to add or change AI models, run the checks again, or remove everything the setup created.

The gateway runs in your Azure subscription, in front of Azure AI Foundry and the models you choose: GPT for text, Claude if you want it, and an image model for web parts that create images. CanoeUX never sees your data, and the web parts don't collect it. The Azure and AI services you deploy are billed to your subscription, under their own terms.

CNX Azure AI Setup web part walkthrough (5:48). Captions are available in the player.

Features

One setup for every web part

The gateway's address is saved once for your whole organization, so every CNX web part that uses AI finds it on its own. Readers of the page simply see whether AI is ready and which models are available; page editors see the full setup.

An Information Technology site page with a SharePoint AI Gateway banner and the CNX Azure AI Setup web part below it. The web part reports “Connected and ready. Text models: claude-opus-5-5, gpt-6.1-sol.”, lists six passed checks, from Gateway reachable to Image model available, with a Run checks again button, and lists the models web parts can use: claude-opus-5-5, gpt-6.1-sol and gpt-image-2.5-sunburst.

View full-size image (opens in a new tab)

Your own gateway, in your own Azure subscription

The SharePoint AI Gateway is an Azure Function protected by Microsoft Entra ID: only signed-in people in your organization can use it. It keeps its secrets in Azure Key Vault, reaches Azure AI Foundry with a managed identity, limits how many AI requests each person can make per hour, and doesn't store or log the prompts, images or content it receives.

Diagram of the SharePoint AI Gateway. SharePoint and Teams in Microsoft 365 send AI requests (1) to the SharePoint AI Gateway, an Entra-protected Azure Function. The gateway retrieves secrets from Azure Key Vault with its managed identity (2) and sends authenticated AI requests over HTTPS (3) to Microsoft Foundry, which hosts GPT, Claude, image and other models. Microsoft Entra ID authenticates users and the app registration that protects the gateway with delegated tokens.

View full-size image (opens in a new tab)

Guided deployment

Deploy to Azure opens a guided form in the Azure portal, with the values it needs shown on the page, ready to copy. Prefer the command line? Run one command in Azure Cloud Shell instead. No Azure rights? Send setup request to your Azure admin emails them the link, the values and the remaining steps.

Checks that explain what to fix

After you connect, the web part checks that the gateway answers, that sign-in works, that Key Vault is reachable and that models are deployed, and explains how to fix anything that fails. Select Run checks again at any time.

The newest models, your choice

The setup deploys GPT-6.1 Sol for text by default, Claude Opus 5.5 if you want it (with your own Anthropic API key or deployed in Azure), and GPT-image-2.5 Sunburst for images. Any other GPT, Claude or image model you deploy later on the same Azure AI Foundry resource is found automatically, with no redeployment.

Clean removal

When you no longer need AI, one Cloud Shell command deletes everything the setup created in Azure and Microsoft Entra ID, and Disconnect all web parts removes the connection from SharePoint.

Configuration

After the CanoeUX SharePoint Web Parts Essentials package has been deployed to your SharePoint tenant, add CNX Azure AI Setup to a SharePoint page, then follow its three steps to connect Azure AI. You need an Azure subscription where you have Owner rights, and a SharePoint Administrator to connect the web parts.

  1. Step 1: Create a page for the setup

    Create a SharePoint page to host the web part. You can use any site where you have permission to edit pages, but a site for administrators is recommended. You'll come back to this page to check the gateway and manage the AI models.

    Important: Everyone who can open the page can see whether AI is ready. Only page editors see the setup, and only a SharePoint Administrator can connect or disconnect the web parts.

  2. Step 2: Add the web part

    1. Select Edit in the upper-right corner of the page.
    2. Hover over the section where you want the web part to appear, and select + Add a new web part.
    3. Search for CNX, and then select CNX Azure AI Setup.

    The web part shows its three steps: Deploy the gateway to Azure, Connect all web parts and Check the connection, followed by Remove the gateway.

  3. Step 3: Deploy the gateway to Azure

    In step 1, choose how to deploy the gateway.

    CNX Azure AI Setup before the gateway is connected. Step 1, Deploy the gateway to Azure, offers Deploy to Azure (recommended), PowerShell script and I already have a gateway. With Deploy to Azure selected, it explains what to do when the Azure portal opens, shows the SharePoint URL to paste with a Copy button, a Deploy to Azure button and a Send setup request to your Azure admin link. Steps 2, Connect all web parts, and 3, Check the connection, follow.

    View full-size image (opens in a new tab)

    Available options

    Deploy to Azure (recommended)

    Copy the SharePoint URL shown on the page, then select Deploy to Azure. The Azure portal opens in a new tab. If it asks Do you trust the authors code?, select Yes, I trust the authors to show the form.

    The Azure portal's “Do you trust the authors code?” prompt, with the Yes, I trust the authors and No, I don't trust the authors buttons.
    PowerShell script

    Prefer to review the deployment first? Copy the command shown and run it in Azure Cloud Shell (Bash or PowerShell), signed in with Owner rights on the subscription. Or select Download the script, review it, and run it from your own scripts or pipelines.

    Step 1 with PowerShell script selected: a Cloud Shell command that downloads and runs Deploy-SharePointAiGateway.ps1 with the site's SharePoint URL, a Copy button, and the Open Azure Cloud Shell and Download the script links. Below the steps, Remove the gateway shows the command that deletes everything the setup created.

    View full-size image (opens in a new tab)

    I already have a gateway

    If the gateway is already deployed, skip to step 2. You'll need its Function URL.

    Send setup request to your Azure admin

    If you don't have Azure rights, select Send setup request to your Azure admin. They receive the link, the SharePoint URL, the command and the remaining steps by email.

  4. Step 4: Fill in the Azure form

    The Set up the SharePoint AI Gateway form has one Settings tab.

    The Set up the SharePoint AI Gateway form in the Azure portal: a description of what it deploys, then Subscription, Region, SharePoint URL, Environment, Resource group (Create new or Use existing) and Name, followed by the start of the Microsoft Entra app registration section.

    Available options

    Subscription, region and SharePoint

    Choose your Subscription and the Region for the gateway's resources, then paste the SharePoint URL you copied from the setup page.

    Resource group

    Choose an Environment, then Create new or Use existing. For a new resource group, leave Name empty and it's named for you, for example rg-spaigateway-prod-cac.

    Microsoft Entra app registration

    The Azure portal can't create the app registration that protects the gateway, so the form gives you a command for it. You need the Application Administrator, Cloud Application Administrator or Global Administrator role.

    1. Select Open Azure Cloud Shell, and copy the Command from the form.
    2. Paste the command into Cloud Shell, press Enter, and wait for it to finish.
    3. Copy the Application (client) ID it prints, and paste it into the form.
    Azure Cloud Shell after running the command with -AppRegistrationOnly: the script prints the Application (client) ID and asks you to paste it into the Application (client) ID box of the Azure portal form.

    View full-size image (opens in a new tab)

    Tip: The command also lets SharePoint web parts sign in to the gateway, so there's no API access request to approve in the SharePoint admin center.

    AI region and models
    The rest of the Azure portal form: the Microsoft Entra app registration instructions with an Open Azure Cloud Shell link, the Command to run and the Application (client) ID box, then AI region set to East US 2, GPT model set to GPT-6.1 Sol (recommended) with a capacity of 50, Claude set to Don't use, Image model set to GPT-image-2.5 Sunburst (recommended), and a usage limit of 30 AI requests per person per hour.
    • AI region: where AI requests are processed. It can differ from the region of the other resources. East US 2 and Sweden Central offer the newest GPT, Claude and image models.
    • GPT model: GPT-6.1 Sol is recommended. The form checks your subscription's quota and uses the deployment type that has room.
    • Claude: Don't use, Use my Anthropic API key (stored in Key Vault) or Deploy in Azure.
    • Image model (optional): GPT-image-2.5 Sunburst for quality, GPT-image-2.5 Flare for speed, GPT-image-2 for high resolution, or Don't deploy if no web part will create images.
    • Usage limit: how many AI requests each person can make per hour (30 by default).
    Review and create

    Select Review + create, check the configuration, and start the deployment. It normally takes a few minutes.

    The Overview page of the deployment in the Azure portal: “Your deployment is complete”, with the two sharepoint-ai-gateway-prod deployments and the rg-spaigateway-prod-cac resource group created successfully.

    View full-size image (opens in a new tab)

  5. Step 5: Connect all web parts

    1. When Azure reports that the deployment is complete, select Outputs in the deployment's navigation.
    2. Copy the value of functionUrl, at the top of the list.
    3. Back on the SharePoint page, paste it into step 2, and select Connect all web parts.
    The Outputs page of the deployment, listing functionUrl, nextSteps, resourceGroup, apiAppName and apiAppId, each with a Copy button. The functionUrl is first.

    View full-size image (opens in a new tab)

    Step 2, Connect all web parts, with the Function URL pasted into the box and the Connect all web parts button, which needs the SharePoint Administrator role.

    View full-size image (opens in a new tab)

    Important: Connecting needs the SharePoint Administrator role, because the connection is saved for your whole organization. Without it, the page shows a PnP PowerShell command to pass on to a SharePoint administrator. The web part checks the address first, so a wrong one is never saved.

  6. Step 6: Check the connection

    After you connect, the web part checks the gateway, sign-in, Key Vault and the models, and lists what passed and what needs attention, with how to fix it. It then lists every model the web parts can use, with its provider and whether it writes text or creates images.

    The connected gateway: its Function URL with a Copy button, the gateway version, six passed checks (Gateway reachable, Sign-in works, Key Vault accessible, Models deployed, Claude available and Image model available), the Run checks again button, and the models web parts can use.

    View full-size image (opens in a new tab)

  7. Step 7: Add or change models

    Come back to this page whenever you need to change the GPT model, add Claude, add an image model or deploy any other model. Add or change models shows the values the Azure form asks for, ready to copy, and step-by-step guides for each change. Models you deploy on the gateway's Azure AI Foundry resource appear automatically.

    The Add or change models section: the Azure details the Azure form asks for, from Subscription ID to GPT model, each with a Copy button; collapsed guides on how to change the GPT model, add Claude, add an image model or add any other model; the commands to list and deploy models with Deploy to Azure and Open Azure Cloud Shell; then Remove the gateway with its command and the Disconnect all web parts button.

    View full-size image (opens in a new tab)

  8. Step 8: Publish the page

    Select Publish (or Republish) in the upper-right corner. The page stays your control point for AI: come back to it to run the checks again, review the models, or remove the gateway.

Limits

Requirements

  • An Azure subscription where you have Owner rights, and quota for the models you choose in the AI region.
  • The Application Administrator, Cloud Application Administrator or Global Administrator role for the app registration.
  • The SharePoint Administrator role to connect or disconnect the web parts.

Costs and usage

  • The gateway, Azure AI Foundry and the models run in your subscription and are billed by Microsoft (or by Anthropic, when you use your own API key), under their own terms.
  • Azure never deletes a model when you switch to another: the old one costs nothing unless it's used, but keeps its quota until you delete it.
  • The hourly limit applies to each person, and counts every AI request from every web part.

Remove the gateway

Run the command shown under Remove the gateway in Azure Cloud Shell. It lists what it will delete and asks you to type DELETE first, then removes everything the setup created, including earlier CNX Icon Studio AI setups. Then select Disconnect all web parts on the page.

Examples

  • IT administration site

    Keep the setup on a page of your IT or Microsoft 365 administration site, where administrators check the gateway and manage the models.

  • A pilot before a wider rollout

    Deploy the gateway with a low usage limit, turn on AI in a few web parts on pilot sites, then raise the limit when you're ready.

Best practices

  • Choose an AI region close to your users that offers the models you need, and check your subscription's quota before you deploy.
  • Keep the usage limit modest at first, and review your Azure costs after the first weeks.
  • Run the checks again after any change in Azure, and before you turn on AI in more web parts.
  • Tell people which web parts use AI, and that they should review AI results before using them.
  • Delete models you no longer use, so they don't hold your quota.

Release notes

  1. Initial release

    November 2026

    • First release, included in CanoeUX SharePoint Web Parts Essentials.

Support

Have a question about CNX Azure AI Setup, or found something that doesn't work as expected? Contact CanoeUX and tell us what you expected to happen and what happened instead. Screenshots help, and for display or access issues, so does the output of CNX User Diagnostics.

Planning a wider intranet? Explore our SharePoint and digital workplace services.

Free of charge

Support CanoeUX

CanoeUX SharePoint Web Parts Essentials is provided free of charge.

If these web parts helped you save time or deliver value to your organization, consider supporting future development.