Skip to main content
Version: 0.0.41

Suggestions

The Suggestions tab shows example prompts generated for this data source during semantic extraction. Suggestions serve as starter questions that guide users toward meaningful queries for the data.

Where Suggestions Come From​

Suggestions are generated automatically by LLM agents during the semantic extraction process. When semantic extraction runs, background AI agents analyze the semantic layer — the schema, tables, column descriptions, data types, and hierarchies — and generate relevant example questions based on what the data contains and how it is structured.

This means:

  • Most suggestions are produced automatically without any manual input from the Builder
  • The quality and relevance of suggestions depends on how well the data source has been described (column names, table descriptions, synonyms)
  • A data source with rich, well-written descriptions will produce better suggestions than one with sparse or technical-only naming
note

Language models are creative and sometimes unpredictable. Not all generated suggestions will be semantically correct or useful for your specific use case. Some may be too generic, misinterpret relationships between tables, or suggest analyses that do not reflect actual business questions. Always review suggestions after extraction.

Each Suggestion Has​

  • Title — a short label displayed to the user as a clickable chip
  • Prompt — the full question text sent to the AI when the suggestion is clicked

Managing Suggestions​

Evaluating Suggestions​

After semantic extraction completes, review the generated suggestions and assess their quality:

  • Does the question make sense given what this data source contains?
  • Is the phrasing clear and useful for your end users?
  • Does the question reflect actual business questions, not just schema-level observations?

Remove suggestions that are misleading, technically incorrect, or not relevant to your users' needs.

Removing Suggestions​

Click the delete action on any suggestion to remove it.

Locking Suggestions​

Lock a suggestion to protect it from being overwritten the next time semantic extraction runs. Manually crafted or reviewed suggestions should be locked to preserve them.

Adding Custom Suggestions​

You can add your own suggestions manually to complement or replace the auto-generated ones:

  1. Click Add Suggestion.
  2. Enter a title (short label shown as a chip) and a prompt (the full question text sent to the AI).
  3. Save.

Custom suggestions are a good way to highlight the most important or frequently needed queries for a data source.

Display Views​

Toggle between List View and Card View to browse suggestions in different layouts.

Integration with Custom Models​

Suggestions from a data source are surfaced to users only when the Custom Model linked to this data source has Show Datasource Suggestions enabled in its General tab configuration.