Skip to main content

Scalar Functions Overview

Scalar functions in FlockMTL operate on data row-by-row, providing powerful operations for text processing, embeddings, image analysis, and machine learning tasks directly within SQL queries.

1. Available Functions​

  • llm_complete: Generates text completions based on model and prompt, supports both text and image inputs

  • llm_filter: Filters rows based on a prompt and returns boolean values, supports both text and image inputs

  • llm_embedding: Generates vector embeddings for text data, used for similarity search and machine learning tasks (text only)

2. Function Characteristics​

  • Applied row-by-row to table data
  • Supports text generation, filtering, and embeddings
  • Multimodal support: Process both text and image data using the same functions
  • Context columns: New API design using context_columns arrays for flexible data input

3. Context Columns API​

All scalar functions now use a unified context_columns API structure:

'context_columns': [
  {'data': column_name},                    -- Basic text column
  {'data': column_name, 'name': 'alias'},   -- Text column with custom name
  {'data': image_url, 'type': 'image'}      -- Image column
]

Each context column supports three properties:

  • data (required): The SQL column data
  • name (optional): Custom name for referencing in prompts
  • type (optional): Data type - "tabular" (default) or "image"

4. Use Cases​

  • Text generation and content creation
  • Dynamic filtering with visual and textual criteria
  • Semantic text and image representation
  • Machine learning feature creation
  • Multimodal analysis combining text and visual data
  • Content moderation for both text and images