Use of general-purpose AI models in Adverity¶
This page summarizes how general-purpose AI models are used in Adverity products following Article 53 of the EU AI Act.
Introduction¶
Users have access to general-purpose AI models in Adverity only as part of the offered features. The models are not intended to be integrated into other systems or used separately. AI is used in the following features:
Transformations Copilot¶
Transformations Copilot uses the OpenAI GPT-5.4 model to generate responses. The model is used without any fine-tuning, keeping the original parameters. For more information about the model, see the OpenAI documentation. OpenAI documents the required details of their models according to their statement.
When using OpenAI models through Adverity, OpenAI’s Services Agreement applies.
The Transformations Copilot processes only the prompts submitted by users. The Transformations Copilot does not have access to data stored in Adverity, such as data extracts and user details.
The Transformations Copilot accepts only text prompts and returns text responses.
The Transformations Copilot is available to all Adverity users starting from the release 2024.16.
Use policies¶
Transformations Copilot is intended to provide information only for the following tasks, and only this usage is acceptable:
Choosing the best-suited instruction for your transformation request.
Configuring the selected custom script instruction.
Creating and configuring Python and RegEx expressions to configure your custom script transformation.
Ask your dashboard¶
Ask your dashboard feature uses the OpenAI GPT-5 model to generate SQL queries and responses. The model is used without any fine-tuning, keeping the original parameters. For more information about the model, see the OpenAI documentation. OpenAI documents the required details of their models according to their statement.
When using OpenAI models through Adverity, OpenAI’s Services Agreement applies.
Only column names and necessary database metadata from the dashboard’s widgets are sent to OpenAI to generate SQL queries, which are then executed within Adverity’s secure environment. Your actual data remains within Adverity, and no personal data is shared with external providers. The summary statistics of your data may be shared with OpenAI if it is requested in your prompt, however, the dataset itself remains in Adverity. Data access is strictly scoped to the data behind the widgets on the current dashboard.
Note
For customers who require their data to be processed and stored exclusively within the European Union (EU), Adverity uses Microsoft’s Azure OpenAI Service. This service is hosted in the EU and complies with all EU data residency requirements.
Ask your dashboard feature accepts only text prompts and return responses as text, tables, and bar and line chart images visualizing data.
Ask your dashboard feature is available as a premium feature starting from the release 2026.16. This feature is only available on instances using Snowflake Managed by Adverity.
Use policies¶
Ask your dashboard feature is intended to provide information only for the following tasks, and only this usage is acceptable:
Gaining immediate insights into dashboard data without manual querying.
Asking follow-up questions about the data shown on a specific dashboard.
AI Insights Widget¶
The AI Insights Widget uses the Gemini 3.1 Pro model via the Google Vertex AI API to generate written analysis from dashboard data. The model is used without any fine-tuning, keeping the original parameters. For more information about the model, see the Google Vertex AI documentation.
When using Google models through Adverity, Google’s Cloud Terms of Service applies.
The underlying data from the dashboard widgets you select, or visual screenshots of your dashboard pages, are sent to Google Gemini to generate the analysis. Google does not use customer data to train or fine-tune its models without explicit customer permission. For information about data residency, see the Vertex AI data residency documentation.
Note
The Gemini 3.1 Pro model is currently available at global endpoints only, meaning data may be processed outside the European Union. The AI Insights Widget is not enabled for customers whose contracts prohibit AI processing or restrict data processing to EU locations only.
The AI Insights Widget accepts text prompts and dashboard widget data selections, and returns responses as formatted text.
The AI Insights Widget is available starting from the release 2026.16.
Use policies¶
The AI Insights Widget is intended to provide information only for the following tasks, and only this usage is acceptable:
Generating written summaries and analysis from dashboard data.
Identifying changes, anomalies, and trends in dashboard data.
Producing audience-tailored written insights for reporting purposes.
Explore Copilot¶
Explore Copilot uses Google Gemini models via the Google Vertex AI API to build and edit dashboard widgets from natural language prompts. The models are used without any fine-tuning, keeping the original parameters. For more information about the models, see the Google Vertex AI documentation.
When using Google models through Adverity, Google’s Cloud Terms of Service applies.
To build a widget, your prompt, the names and metadata of the metrics and dimensions available in your workspace, the dashboard theme, and any images you attach are sent to Google Gemini. Your actual data remains within Adverity. Google does not use customer data to train or fine-tune its models without explicit customer permission. For information about data residency, see the Vertex AI data residency documentation.
Note
The Gemini models are currently available at global endpoints only, meaning data may be processed outside the European Union. Explore Copilot is not enabled for customers whose contracts prohibit AI processing or restrict data processing to EU locations only.
Explore Copilot accepts text prompts and images, and returns a configured widget.
Explore Copilot is currently in Beta and is available starting from the release 2026.27.
Use policies¶
Explore Copilot is intended to provide information only for the following tasks, and only this usage is acceptable:
Building and editing dashboard widgets from natural language prompts.
Choosing chart types and configuring metrics, dimensions, filters, and styling for a widget.
Connect Agent¶
Connect Agent uses the OpenAI GPT-6 Luna model to answer questions about your workspace and to make changes that you approve. The model is used without any fine-tuning, keeping the original parameters. For more information about the model, see the OpenAI documentation. OpenAI documents the required details of their models according to their statement.
When using OpenAI models through Adverity, OpenAI’s Services Agreement applies.
To answer a request, Connect Agent sends the following to OpenAI:
Your messages and the recent conversation, including the tools Connect Agent used earlier, with their arguments and results.
Your username (your email address), the URL of your instance, the name and ID of your workspace, today’s date, and instructions that depend on what your role allows.
The address of the page you are on, and the names and IDs of the item open on it and of the datastreams you have selected.
The results of the tools Connect Agent uses. Depending on your request, these can include the configuration of your datastreams, authorizations, and destinations, rows from your data extracts, the results of queries on the data Adverity has loaded into the Data Warehouse for your workspace, and, for administrators, the names, email addresses, roles, and workspace access of users.
To answer questions about how to use Adverity, Connect Agent looks up this documentation. It sends a question written by the model, which can include details of your workspace, to the Adverity documentation service. The documentation service uses the OpenAI text-embedding-3-small model to search the documentation and the OpenAI GPT-4o mini model to write answers.
Connect Agent acts with the permissions of the user who asks. For each answer, Adverity creates a short-lived key limited to that user’s permissions, the current workspace, and its child workspaces, and deletes it when the answer is complete. Connect Agent applies a change only after the user approves it.
Adverity keeps each user’s current conversation in each workspace until the user starts a new conversation, or until the user or the workspace is deleted. Connect Agent sends its requests to OpenAI with response storage turned off, so OpenAI does not keep them as stored responses. The documentation service keeps the questions it receives for one year.
Note
Connect Agent and the documentation service use OpenAI’s global API, meaning data may be processed outside the European Union. Connect Agent is not available for accounts whose contracts prohibit AI processing.
Connect Agent accepts only text prompts and returns text responses, links to Adverity pages, approval requests for changes, and suggested follow-up actions.
Connect Agent is available starting from the release 2026.43 for accounts where it is enabled.
Use policies¶
Connect Agent is intended to provide information only for the following tasks, and only this usage is acceptable:
Looking up and troubleshooting the configuration and status of your workspace, such as datastreams, fetches, authorizations, and destinations.
Looking up data in your data extracts and in the Data Warehouse.
Making changes in your workspace after you approve them, such as configuration changes, starting fetches, and, for administrators, managing users and notifications.
Answering questions about how to use Adverity.