The COMPLETE function enables natural language generation within Snowflake Cortex. It accepts a prompt and produces a text-based completion using an underlying large language model (LLM). COMPLETE can generate single-turn or multi-turn responses, depending on how the prompt is structured. It is used for tasks such as drafting summaries, generating recommendations, writing structured output such as JSON, or extending existing text. It is not a classification tool-that role is served by CLASSIFY_TEXT-and it does not perform linguistic translation (TRANSLATE) or document extraction (PARSE_DOCUMENT).
COMPLETE allows developers and analysts to embed AI-driven generative capabilities directly within SQL pipelines or Snowpark processes, eliminating the need for external LLM services.