Today, we’re excited to announce Snowflake Cortex AISQL in public preview, bringing powerful AI capabilities directly into Snowflake’s SQL engine. Cortex AISQL enables customers to build scalable AI pipelines across multimodal enterprise data with familiar SQL commands. Process text (public preview), images (public preview) and audio (public preview coming soon) faster and more cost effectively and gain deeper insights from both structured and unstructured data at the same time.

Data analytics is undergoing a profound transformation, driven by AI. Today's enterprises recognize that valuable insights extend far beyond traditional tabular data — they're often embedded within diverse unstructured sources such as documents, images and audio files. 

As data variety grows, analytics requires increasingly sophisticated multistep processes to extract, synthesize and combine insights across sources. However, analyzing this rich tapestry of data at enterprise scale presents significant technical and operational challenges. 

Organizations struggle with stitching together various specialized tools and skill sets to process and analyze this information, compromising efficiency and scalability while resulting in delayed and lower-quality insights.

To address these challenges, Snowflake is integrating powerful generative AI capabilities directly into its core SQL engine, empowering analysts to accomplish more in less time through:

• Expressive and composable AI operators: A new set of AI-powered operators that seamlessly maps with existing SQL primitives such as FILTER and AGGREGATE. This makes it straightforward for analysts to build scalable AI pipelines using familiar SQL commands for detecting subtle anomalies in operational data, aggregating actionable insights from customer call transcripts or classifying extensive image libraries.

• Native support for multimodal data: The introduction of a new FILE data type allows direct referencing of multimodal data (including text, images and audio) within Snowflake tables. All AI operators are designed to work seamlessly across these various data modalities, eliminating the need for separate processing systems.

• Significant performance and cost improvements: With native functionality and performance optimizations built into Snowflake’s SQL engine, Cortex AISQL substantially reduces processing time and computational expense. Internal benchmarks of Cortex AISQL Performance Optimization (private preview) show up to 70% query run time reduction for operations such as FILTER and JOIN,with corresponding cost savings compared to manual implementations, directly addressing concerns about AI scalability and economic viability for enterprise adoption.

"Snowflake Cortex AISQL accelerated our development of our Service Technician application that enables our technicians to easily interact and analyze thousands of user manuals in multiple languages to solve customer problems faster than we could have imagined. The game changer for us has been how Cortex AISQL seamlessly integrates diverse data types, from text to images and beyond, paving a faster path from data to insights that transforms how our teams work," said Ahmad AI-Mashahedi, Lead Data Scientist for Toyota Material Handling Europe.

The core capabilities of Cortex AISQL

Cortex AISQL integrates AI capabilities directly into SQL environments, enabling sophisticated multistep analysis across diverse data types. In the following sections, we'll explore how these core capabilities optimize performance, reduce costs and enable complex analytical workflows.

Unified AI operations for diverse data with Cortex AISQL

Cortex AISQL’s core strength lies in its ability to allow users to construct sophisticated AI pipelines across multiple modalities of data directly within their familiar SQL environment. This integration eliminates the separation between a customer’s data stack and the AI stack, offering a unified platform for sophisticated analysis across a wide range of data.

Cortex AISQL introduces AI operators as native SQL primitives, fully embedded within the Snowflake ecosystem. For example:

• AI_FILTER applies AI-driven filtering logic directly within the WHERE clause. 

• JOIN operations can now be based on AI relationships rather than just predefined keys, enabling connections between different data formats based on semantic understanding (for example, joining resumes to job descriptions based on an AI's evaluation of fit). 

• AI_AGG, a pioneering AI-powered aggregation operator, is designed to work with GROUP BY to derive complex insights from diverse rows of data. 

• Enhanced AI_CLASSIFY can support multilabel classification across both text and images.

Cortex AISQL significantly broadens the addressable data surface, with all new operators engineered to work efficiently across text, image and, soon, audio data, through unified operators. Since there is no need for any additional services, it uplevels data analysts into AI superheroes who can work with all types of data.

Unlocking deep insights: Advanced multistep analytics with AI pipelines

AI capabilities unlock new dimensions of analysis and the ability to tackle increasingly complex multistep questions, going beyond simple lookups. While retrieval-augmented generation (RAG) excels at point-lookup tasks, deeper analytical problems require orchestrating multiple AI-powered steps to synthesize insights.

Answering this involves:

• Searching through large document collections to identify relevant information

• Filtering key criteria from relevant document sections, such as CEO changes and the industry vertical the organization falls under (renewable energy)

• Extracting key information such as revenue figures and outlook statements from unstructured data

• Joining unstructured insights with structured data for year-over-year revenue growth calculation

• Synthesizing the findings into a clear, coherent summary that provides a deep understanding of the trends and future prospects

Cortex AISQL makes this complex analytical workflow achievable through its composable AI operators and native SQL integration. By allowing analysts to chain together AI-powered operations, from document filtering to semantic extraction to intelligent joining, all within familiar SQL syntax, it eliminates the need for multiple specialized tools and custom code. This unified approach transforms what would traditionally require data science expertise and weeks of development into straightforward SQL queries that business analysts can build and modify in minutes.

As we look ahead, these types of complex analytical questions will become increasingly common in business environments. Organizations will need to combine multiple analytical patterns across structured and unstructured data to extract meaningful insights from their data. Cortex AISQL addresses this growing need by allowing analysts to construct sophisticated AI pipelines, enabling teams to answer these composite questions with SQL natively.

Cut query time and cost, but not accuracy

Reducing processing time and cost is vital for enterprise AI. By running AI functions inside Snowflake’s core query engine, we combine AI and structured-data processing in one place, using the Snowflake warehouse’s parallelism and batching to optimize both at once.

This scalability is essential for real enterprise use cases. Consider a task such as “Find modern, attractive houses within 10 miles of San Francisco in my price range.” Our system intelligently plans execution across multiple WHERE clauses: applying standard filters for price and distance to first narrow the data set before using a visual language model to assess "modern and attractive" qualities, ensuring maximum efficiency. 

A core focus of Cortex AISQL is delivering these powerful AI capabilities with exceptional performance and cost efficiency, without compromising the accuracy of insights. We implemented Adaptive LLM Optimization (private preview) strategically to orchestrate smaller models for simpler tasks, while reserving larger LLMs for more complex analysis. This tiered approach ensures optimal resource utilization while maintaining high-quality results. 

Cortex AISQL helps industries from financial services to retail to healthcare unlock new insights and automate complex processes from both their structured and unstructured data.

• Financial services: Capital market institutions can now automate complex corporate action processing with unprecedented efficiency. Using AI_FILTER on vast feeds of news and regulatory documents, analysts can instantly identify relevant event announcements. These findings are then seamlessly joined with internal holdings positions through AI-powered intelligent matching of organization names and identifiers, enabling precise impact assessment and informing next-step actions that were previously impossible at scale.

• Retail and ecommerce: Customer experience teams can detect emerging product quality issues in real time by using AI_AGG to group and analyze thousands of customer reviews by SKU. The system applies AI_FILTER to flag products showing concerning patterns, such as sudden surges in negative sentiment or specific defect reports, enabling quality assurance teams to respond rapidly to potential issues before they impact brand reputation or trigger costly recalls.

• Healthcare: Medical researchers accelerate discoveries through AI-powered analysis that bridges unstructured clinical notes with structured patient records. By applying AI_FILTER to physician notes, researchers can identify patients with specific symptom patterns not captured in standard diagnostic codes. These insights can then be joined, using JOIN, with lab results and medication histories to uncover subtle correlations that might indicate new treatment pathways or previously unrecognized risk factors.

These use cases demonstrate how Cortex AISQL transforms slow, manual processes into highly efficient, intelligent workflows. 

Cortex AISQL availability

Transform data analytics today

Cortex AISQL is redefining how businesses go from their data to insights. With easy connection to all of a customer’s data sources across modalities via Openflow, Snowflake is making it easier than ever to use AI-powered SQL operators to analyze structured and unstructured data together. Cortex AISQL delivers high-performance processing at a lower cost than manually orchestrated AI pipelines, enabling trusted insights across the enterprise while maintaining the security and governance capabilities Snowflake is known for.

The public preview of Cortex AISQL is now open for all Snowflake customers. Get started today and experience how these powerful capabilities can transform data analytics. 

Get started with Cortex AISQL today with the following resources: 

• Try Cortex AISQL

• Watch the demo: See Cortex AISQL in action

  Documentation: Get started with Cortex AISQL today

Forward Looking Statements

This article contains forward-looking statements, including about our future product offerings, and are not commitments to deliver any product offerings. Actual results and offerings may differ and are subject to known and unknown risk and uncertainties. See our latest 10-Q for more information.