Skip to main content

AI Tables

Intelligence beyond SQL—use AI to classify, extract, and analyze data in ways traditional queries can't.

What are AI Tables?​

AI Tables are special Views where large language models add intelligent columns to your data. While regular Views use SQL for transformations, AI Tables use AI to perform tasks like categorizing text, extracting entities, analyzing sentiment, or scoring quality—things SQL simply can't do.

They're represented as purple nodes in your graph.

AI Table Node

Purple AI Table nodes stand out in your pipeline

Why AI Tables Matter​

Handle unstructured data: Analyze text, descriptions, comments, and free-form fields.

Intelligent categorization: Classify expenses, tickets, or leads without manually defining rules.

Extract structure from chaos: Pull structured information from messy text.

Human-like reasoning: Get AI judgments on quality, sentiment, or intent.

Flexible taxonomy: Define categories naturally—no need for regex or keyword lists.

How AI Tables Work​

Manual Refresh Design​

Unlike regular Views that update automatically, AI Tables require manual refresh. This gives you control over when to spend AI tokens.

  1. AI creates an AI Table with new columns (initially empty/NULL)
  2. You review the setup and prompt
  3. You click "Refresh" when ready to process
  4. AI processes each row and fills in the new columns
  5. Check results and iterate if needed

AI Table Before/After Refresh

Before refresh: NULL columns. After refresh: AI-generated values

Creating an AI Table​

Just ask naturally:

@[customer_feedback] Classify each comment as Positive, Negative,
or Neutral and extract the main topic mentioned

The AI will:

  1. Create a View referencing your data
  2. Add new columns for classifications
  3. Generate a prompt for the LLM
  4. Wait for you to trigger the refresh

Common Use Cases​

Expense categorization: Turn messy expense descriptions into clean categories

  • Input: "Taxi to airport"
  • Output: Category = "Transportation", Subcategory = "Ground Transit"

Customer feedback analysis: Understand themes in support tickets

  • Input: "App keeps crashing when I upload photos"
  • Output: Sentiment = "Negative", Issue = "Technical - App Stability"

Lead scoring: Evaluate sales opportunities based on descriptions

  • Input: "Mid-market company, interested in enterprise plan, urgent need"
  • Output: Priority = "High", Likelihood = "Strong", Urgency = "Immediate"

Entity extraction: Pull structured data from text

Taxonomy mapping: Organize products into hierarchical categories

  • Input: "Organic Free-Range Eggs, Large"
  • Output: L1 = "Groceries", L2 = "Dairy & Eggs", L3 = "Eggs", Attributes = ["Organic", "Free-Range"]

Quality scoring: Rate content or data quality

  • Input: Customer review text
  • Output: Helpfulness_Score = 8, Detail_Level = "High", Actionability = "Specific"

AI Table Examples

Common AI Table use cases: classification, extraction, scoring

Tips & Best Practices​

Filter before processing: If you have 100,000 rows but only need to analyze recent feedback, filter first. Saves tokens and time.

Be specific in prompts: "Classify as Positive/Negative/Neutral" is better than "Analyze sentiment"

Request confidence scores: Ask for confidence percentages to identify uncertain classifications

Use hierarchical categories: Request multi-level taxonomy (Category → Subcategory → Detail)

Include reasoning columns: Add a "reasoning" column so you can see why the AI made each decision

Iterate on prompts: If results aren't right, edit the prompt and refresh again

Consider costs: Large AI Tables can consume significant tokens. Start small to test.

Understanding Token Usage​

AI Tables show token consumption estimates before processing. You'll see:

  • Estimated tokens per row
  • Total token estimate
  • Model being used

This helps you decide whether to filter more or adjust the prompt.

Token Usage Display

Token usage estimates before processing

Model Selection​

You can choose which AI model to use:

  • Faster models: Quick results, good for simple classifications
  • Smarter models: Better reasoning for complex tasks

Right-click the AI Table to select your model preference.

Refreshing and Updating​

Manual refresh only: Click the refresh button to process

Partial updates: Can refresh just new rows in some cases

Re-run anytime: Change the prompt and refresh to reprocess

Cost visibility: Always see token usage before committing