Overview

​

1. Available Functions

• llm_complete: Generates completions — text, image, and voice (audio)
• llm_filter: Filters rows — text, image, and voice (audio)
• llm_embedding: Text embeddings (image/voice via describe or transcribe first)

​

2. Function Characteristics

• Applied row-by-row to table data
• Modalities: Each LLM function page documents Text, Image, and Voice support with examples
• Context columns: Unified context_columns array for tabular, image, and audio inputs
• Structured output: Schema-shaped JSON via model_parameters where applicable (see each function’s Output)

​

3. Context Columns API

'context_columns': [
  {'data': column_name},                    -- Text (default type: tabular)
  {'data': column_name, 'name': 'alias'},   -- Text with prompt alias
  {'data': image_url, 'type': 'image'},     -- Image
  {'data': audio_path, 'type': 'audio', 'transcription_model': 'whisper-1'}  -- Voice
]

• data (required): SQL column value
• name (optional): Alias for the prompt
• type (optional): "tabular" (default), "image", or "audio"
• transcription_model (required for audio): e.g. whisper-1 (OpenAI / Azure)

​

4. Common Use Cases

• Text generation, classification, and filtering
• Image description and visual filtering
• Voice transcription and call analytics
• Embeddings and similarity search (text; image/voice via two-step workflows)