Copilot for Real-Time Intelligence

Copilot in the Fabric Real-Time Intelligence workload is an AI assistant that helps you query, analyze, and explore your real-time data. Copilot translates natural language into Kusto Query Language (KQL) queries, generates dashboards, and enables interactive data exploration, all without requiring KQL expertise.

Copilot in KQL querysets

Copilot in KQL querysets transforms natural language questions into KQL queries. Describe your data analysis needs in plain language, and Copilot generates the corresponding query. Copilot supports conversational interactions, so you can refine queries and ask follow-up questions without starting over.

For details on how to use Copilot in KQL querysets, see Copilot for writing KQL queries.

Copilot in Real-Time Dashboards

Copilot in Real-Time Dashboards simplifies dashboard creation and data exploration:

• Generate dashboards: Select a data table in Real-Time Hub or a KQL queryset and use Copilot to automatically generate a Real-Time Dashboard with an insights page and a data profile page. For details, see Generate a Real-Time Dashboard using Copilot.
• Edit tile queries: Use Copilot to author or modify the KQL query behind a dashboard tile directly in the editing pane, using natural language instead of writing KQL manually.
• Explore data interactively: In view mode, use Copilot to ask questions about your dashboard data, filter results, and save insights as new tiles. For details, see Copilot-assisted real-time data exploration.

Copilot for Azure Data Explorer

Copilot also supports Azure Data Explorer (ADX) clusters. When connected to an ADX cluster, Copilot generates KQL queries and explores data in the same way it does for an Eventhouse. A Fabric-enabled capacity is required.

For more information on connecting to ADX from Fabric, see Consume ADX data in Fabric.

Best practices for Copilot KQL queries

The following tips apply to Copilot in both KQL querysets and Real-Time Dashboards:

• Start with simple natural language prompts to learn current capabilities and limitations. Gradually proceed to more complex prompts.

• State the task precisely and avoid ambiguity. Imagine sharing the prompt with a KQL expert without adding oral instructions. Would they generate the correct query?

• Supply relevant information to help the model. Specify tables, operators, or functions that are critical to the query when possible.

• Prepare your database:

  • Add docstring properties to describe common tables and columns. This step is critical for tables or columns with nonmeaningful names.
  • You don't have to add docstrings to tables or columns that are rarely used.
  • For more information, see alter table column-docstrings command.

• To improve Copilot results, select the like or dislike icon to submit feedback.

  Note

  The Submit feedback form submits the name of the database, its URL, the KQL query generated by Copilot, and any free text response you include. Results of the executed KQL query aren't sent.

Improve Copilot accuracy with Private Shots

Copilot enhances prompts by using the most relevant examples (referred to as natural language and KQL pairs, or "shots") from a Public Shots database. This database is curated by the Real-Time Intelligence team, derived from KQL documentation, and available to all Copilot users. The Public Shots database provides a solid foundation but is generic and lacks domain-specific knowledge of your KQL database.

To improve Copilot's ability to generate accurate and complex KQL queries for your specific scenarios, create a Private Shots database.

This approach lets you include advanced KQL queries that address your team's unique requirements. For example, queries that use: - graph semantics, - time series analysis, - anomaly detection, - or stored functions defined in your KQL database.

Private Shots are automatically published from both KQL querysets and Real-Time Dashboards. When you save these artifacts, the KQL queries they contain are published to the Private Shots database, improving Copilot's ability to generate queries that align with your data and use cases.

Limitations

The following limitations apply to Copilot across Real-Time Intelligence:

• Copilot can't modify existing KQL queries in the query editor. If you ask the Copilot chat pane to edit a specific part of an existing query, it doesn't work. However, Copilot understands previous inputs in the chat pane, so you can iterate on queries that Copilot generated before insertion.
• Copilot might produce inaccurate results when the intent is to evaluate data. Copilot only has access to the database schema and doesn't have access to the data itself.
• Copilot responses can include inaccurate or low-quality content. Review outputs before using them in your work.
• People who can meaningfully evaluate the content's accuracy and appropriateness should review the outputs.
• The Copilot chat pane in KQL databases isn't available when Private Link is enabled and Public Access is disabled in the tenant setting.

Responsible AI

To view Microsoft's guidelines for responsible AI in Real-Time Intelligence, see Privacy, security, and responsible use of Copilot for Real-Time Intelligence.

Microsoft is committed to ensuring that AI principles and the Responsible AI Standard guide the AI systems. These principles include empowering customers to use these systems effectively and in line with their intended uses.

Related content

• Copilot for writing KQL queries
• Generate a Real-Time Dashboard using Copilot
• Copilot-assisted real-time data exploration
• Overview of Copilot in Fabric