AI Summaries

What Are AI Summaries?

Every component in your Power Platform environment has metadata — the raw definition of what it is and how it behaves. A plugin step has an assembly name, message, stage, and filtering attributes. A cloud flow has a JSON definition with triggers, conditions, and actions. A form has XML layout with sections, tabs, and controls.

AI Summaries take that metadata and use a large language model to produce a human-readable paragraph explaining what the component does, when it runs, and what it affects. Instead of reading raw XML or JSON, your team reads something like:

How They Work

The summary generation process follows these steps:

1. 1

   Metadata extraction

   Cartographer reads the component's full metadata from Dataverse — schema definitions, form XML, flow JSON, plugin registration details, etc.

2. 2

   Prompt construction

   A tailored prompt is built for the specific component type, including the extracted metadata and instructions for the AI model.

3. 3

   Token management

   If the metadata exceeds the model's context window, Cartographer intelligently truncates it — keeping the most important fields and removing redundant data.

4. 4

   AI generation

   The prompt is sent to the configured AI endpoint (BYOK or Hosted). The model generates a concise summary.

5. 5

   Storage

   The summary is stored on the Component record in Dataverse. It persists across sessions and does not need to be regenerated unless the component changes.

Generating Summaries

There are two ways to generate AI summaries:

Individual component

1. Navigate to the Component Dashboard
2. Click on any component to open its detail panel
3. Click the Generate AI Summary button
4. Wait a few seconds for the summary to appear

Bulk generation

1. In the Component Dashboard, select multiple components using the checkboxes
2. Click Generate Summaries in the command bar
3. Summaries are generated sequentially — a progress indicator shows completion

Token Management and Truncation

Some components have very large metadata — a complex form with 50 tabs, or a cloud flow with hundreds of actions. The AI model has a limited context window, so Cartographer uses intelligent truncation:

• Key identifiers (name, display name, type) are always preserved
• Structural information (form sections, flow actions) is prioritized
• Redundant or boilerplate content (default values, empty fields) is removed
• If content still exceeds the limit, it is truncated with an indicator

Cost Estimates for BYOK

When using BYOK, you pay OpenAI directly for token usage. Here are approximate costs per component summary based on typical metadata sizes:

BYOK vs Hosted: Which Should I Choose?