Skip to main content

Views

SQL transformations the AI creates for you automatically—no SQL writing required.

What are Views?​

Views are the building blocks of your analysis. When you ask the AI a question, it breaks down the work into discrete SQL transformations called Views. Each View is a step in your analytical pipeline—filtering, joining, calculating, aggregating, or reshaping data.

You never have to write SQL yourself, but you can always inspect the code to verify the logic.

View Nodes

Color-coded View nodes showing your analytical pipeline

Why Views Matter​

Transparency: See exactly what SQL the AI wrote. No black-box magic.

Reusability: Reference Views in future questions. Build analysis layer by layer.

Modularity: Complex analysis becomes a chain of simple, understandable steps.

Debugging: If something looks wrong, inspect the View where it happened.

Understanding View Types​

The AI classifies Views by operation. Each type gets a unique color and icon:

Filter - Narrow down rows with WHERE conditions

Convert Types - Cast data types or parse dates

Calculate Columns - Add new computed columns

Aggregate - GROUP BY operations and summaries

Combine - Joins, unions, and set operations

Calculate Groupwise - Window functions and rankings

Reshape - Pivot, unpivot, or unnest arrays

Present - Final formatting for output

Recursion - Recursive queries for hierarchical data

View Types Legend

Each View type has a distinct color and icon

How Views Work​

Automatic Creation​

When you chat with the AI:

  1. AI analyzes your request
  2. Breaks work into logical steps
  3. Creates one View per step
  4. Shows them as nodes in the graph

You simply approve or reject the Views. The AI handles all SQL generation.

Draft Mode​

New Views appear with a red dot indicator—they're in draft mode. You can:

  • Accept individual Views: Click the checkmark
  • Accept all: Approve multiple Views at once
  • Reject all: Discard and ask the AI to try differently

Draft View

Draft Views with red indicators waiting for approval

Inspecting Views​

Click any View node to see:

  • Data preview: First rows of the result
  • View name: Descriptive name the AI chose
  • Classification badge: What type of operation it performs
  • Brief summary: 4-word explanation of the logic
  • Detailed explanation: Full description in expanded mode
  • Dimensions: Row and column counts
  • SQL code: Toggle to "Code" tab to see the query

View Details Panel

Bottom panel showing data, profile, and SQL code tabs

Common Use Cases​

  • Multi-step analysis: Filter → Join → Calculate → Aggregate chain
  • Incremental refinement: Build on existing Views without starting over
  • Data quality checks: Inspect intermediate Views to verify logic
  • Learning SQL: See how the AI writes queries for your requests
  • Reusable components: Create Views once, reference in multiple analyses

Tips & Best Practices​

Accept incrementally: Approve Views as they're created instead of batching at the end. Helps you catch issues early.

Check intermediate steps: Don't just look at the final result. Click through Views to verify each step makes sense.

Name Views meaningfully: You can rename Views. Use names like "high_value_customers" instead of "view_7".

Build on existing Views: Reference Views by name in future prompts: "Using @[monthly_revenue], show the top 5 months"

Understand before accepting: If a View's logic seems unclear, ask the AI to explain it before accepting.

Managing Views​

Rename: Right-click and select "Rename" for clearer names Edit: Right-click and use Quick AI to modify specific Views Delete: Right-click to delete (with dependency warnings) Copy data: Export View results to CSV or copy to clipboard Pop out: Open data preview in a separate window

View Context Menu

Right-click menu for View management

Data Preview Options​

Toggle between three views at the bottom panel:

Data: See the actual rows and columns

Profile: View metadata, statistics, and data quality info

Code: Inspect the SQL query

View Toggle

Switch between Data, Profile, and Code tabs