Automatic text and multimodal content summarization with AI_SUMMARIZE

AI_SUMMARIZE is a managed Cortex AI function that automates the condensation of text and multimodal content. It accepts text inputs or file references in stages (such as images, reports, and contracts) and returns structured summaries with key themes and facts without requiring prompt engineering or complex processing pipelines.

Use cases

Imagine you need to process an entire library of PDF reports or visual diagrams without manually parsing each file.

• Automatically extract main topics and themes from long-form content.
• Summarize complex business documents like contracts and research papers.
• Understand visual content such as diagrams and workflows.
• Process file libraries at scale using file reference tables.

How to use

To summarize text directly in a SQL query:

SELECT AI_SUMMARIZE('These high-end luxury shoes combine excellent craftsmanship with limited comfort...');

What happens in this code:

1. The function receives the long text string as a direct argument.
2. The model processes the content, identifying strengths, weaknesses, and the overall conclusion.
3. The database returns a concise text summary preserving the core meaning.

To summarize files stored in a stage:

SELECT AI_SUMMARIZE(TO_FILE('@my_docs', 'cartoon_image.jpeg')) AS image_summary;

What happens in this code:

1. The TO_FILE object points to the file located in the internal stage.
2. AI_SUMMARIZE reads the image or PDF document multimodally.
3. The database returns a summarized analysis of the file's visual or textual concepts.

Limitations and requirements

• Public Preview feature available to all accounts.
• Inputs of 110 characters or fewer return the original text.
• For aggregations across rows, use the AI_AGG function.
• Format support, file size limits, and context windows vary by region (e.g., AWS Global, Azure Global, GCP Global, or Any Region have specific token and file size thresholds).