Administer AI Builder
Microsoft Power Platform administrators can use the Power Apps admin center and the Power Platform admin center to manage environments and settings for Power Apps and AI Builder. For more information, go to Power Platform administrator guide.
The availability, which is also known as the release status, of AI Builder is dependent on your region. For a breakdown of AI Builder feature availability by region, go to Feature availability by region.
For more information, go to the AI Builder governance whitepaper and select Download.
AI custom model and environment lifecycle
This section applies to AI Builder custom models. It does not apply to prebuilt models.
Lifecycle states of a model
A model can go through different states depending on the makers’ actions. The model states are stored through configurations within the AI Configuration table.
The states are draft, training (transient), trained, publishing (transient), published, unpublishing (transient), training error, importing (transient), and import error.
Application lifecycle management
Makers should be able to continuously update and deploy their models across single or multiple environments.
Managing new versions of a model often requires going through different environments. A typical scenario would be to make model changes in a development environment, qualify the model in a test environment, and predict in a production environment.
In AI Builder, all the environments need to be provided with a Microsoft Dataverse database.
Moving models between environments can be done through the solution concept. Solutions are vehicles to move components between Microsoft Power Platform environments. For more information, go to Introduction to solutions.
For more information on how to distribute an AI model as a solution component, go to Distribute your AI model.
AI Builder models are fully moved, along with user data, during environment backup/restore and environment copy operations.
After restore and copy operations, form processing and object detection models may be in the importing state for a few minutes while copies are made in the back end.
Backup and restore
Microsoft Dataverse has backup and restore capabilities to help protect your apps' data, providing continuous availability of service. System administrators and delegated admin users can use the standard capabilities described in Back up and restore environments.
Backup and restore are fully supported for prediction, object detection, form processing, and prebuilt models.
For object detection and form processing models, the restore process might take some time to be completed. The AI Builder models list shows an "importing" status message while the restore operation is in progress.
For models not supported by backup and restore: If you restore an environment, you'll have to retrain and republish these models to make them available again.
AI Builder consumption model
AI Builder offers a subscription model allowing you to purchase add-ons.
Only certain actions in the product consume credits. The following list is not all-inclusive and preview scenarios don't consume credits.
|AI Builder Studio||Power Apps||Power Automate|
|Train an Object Detection model.
Perform a Quick test on a trained Object Detection and Form Processing model.
Use the Try it out of models with custom documents (appearing in the Get straight to productivity section).
Batch runs of the prediction and trainable category classification models for each row to be predicted.
Scan a business card with the Business Card reader.
Analyze with the form processor.
Detect with the object detector.
+ New image with the text recognizer.
|Run a flow using any of the actions inside the AI Builder category.
Run the generic action Perform a bound action of Dataverse on the entity AI Models and action name Predict.
Each AI Builder model has a different credit consumption mechanism. To perform an assessment, go to the AI Builder calculator.
By default, the credits are unallocated and available as a pool on the tenant. They must be allocated to an environment. If they aren't, users won't be able to consume credits.
This is how administrators stay in control of where AI is used in their organization and, with the role assignments described in Roles and security in AI Builder, who is using it.
As an administrator, you'll assess which environments must consume AI Builder credits. Use the AI Builder calculator to define how many predictions will happen in a monthly period on each one and assess the credits to allocate.
To learn how to allocate credits in the Power Platform admin center, go to Allocate or change capacity in an environment.
As an administrator, you have access to a consumption report that provides the AI credits consumption on a chosen period per environment. This will allow you to fine-tune the credits allocation, which can be updated at any time.
To learn how to download reports, go to Allocate or change capacity in an environment.
Where and how are data stored in Dataverse?
Your AI model is deployed in the region that hosts the environment. For example, if your environment is created in the Europe region, your model is deployed in datacenters in Europe. For more information, go to Environments overview.
Images and documents used for training purposes in object detection and form processing models are persisted in Dataverse. In contrast, images and documents used at prediction time aren't persisted. Examples of non-persisted images and documents are those in a Power Apps component framework (PCF) control and in Power Automate.
Enable or disable AI Builder preview features
Some AI Builder features are released for general availability. Others remain in preview release status.
Preview features appear on the build page with the Preview label. In the Power Platform admin center, administrators control whether users have access to preview features.
By default, the AI Builder preview models feature is enabled for any eligible environment. Eligible environments must have Microsoft Dataverse and be in a supported region. If the environment isn't eligible, the AI Builder preview models feature doesn't appear in the Power Platform admin center.
To control AI Builder preview feature availability:
Sign in to the Power Platform admin center.
In the admin center, go to Environments > [select an environment] > Settings > Features.
On the Features settings page, under AI Builder, enable or disable AI Builder preview models.
Important points related to enabling or disabling the feature
If you disable AI Builder preview models:
- We don't delete existing models that users of this environment have created.
- AI Builder components are disabled.
- Existing experiences that use existing AI Builder components will fail or return errors.
- Admins and owners can delete preview models.
If you enable AI Builder preview models again:
- AI Builder components are available again.
- Components function as they did before the feature was disabled (assuming nothing else has changed).
For more information about enabling or disabling features in the Power Platform admin center, go to Manage feature settings.
Data loss prevention (DLP)
You can control data loss prevention (DLP) policies within Power Platform admin center, Data policies menu item.
Connectors can be listed in three (3) categories: Business, Non-business, and Blocked.
AI Builder is part of the Dataverse connector.
Business and Non-business connectors can’t share data within the same consumption experience in Microsoft Power Platform.
- For example, if you add the Dataverse connector in the Business category, and Microsoft Outlook in the Non-business category, you won’t be able to create a Power Automate flow that gets the output of an AI Builder model and sends it to a recipient in Outlook.
Blocked connectors can’t be used in Power Platform consumption experiences.
To learn more, go to Data loss prevention policies.
Move and copy environments
For prediction and prebuilt models, moving and copying environments is fully supported. For other models, after you move or copy an environment, you have to retrain and republish existing models to make them available again.