Snowflake today announced the public preview of its generative artificial intelligence (genAI) powered SQL assistant — Snowflake Copilot, with limited availability in Amazon Web Services (AWS) US regions. It is designed to make complex SQL queries accessible through simple natural language prompts.

Snowflake Copilot was first introduced in a private preview during last year’s Snowday event. Powered by Snowflake Cortex, Copilot integrates genAI into daily coding tasks with natural language, allowing users to interact with their data using plain text, write SQL queries against relevant data sets, refine queries and filter down insights. Snowflake Cortex is a fully managed service designed to enable organizations to discover, analyze, and build AI apps in the Snowflake Data Cloud more easily.

The development of Copilot benefits from Snowflake's deep understanding of SQL complexities, the vendor touted, as it processes more than 4 billion queries running daily on its Data Cloud platform. This also fuels and fine-tunes Snowflake's text-to-SQL model.

Users can pose data-related questions in plain English, and Copilot will generate the corresponding SQL queries. This feature helps users to write SQL queries faster and analyze data more easily.

Snowflake combines its proprietary SQL generation model with the Mistral Large text generation model to create this text-to-SQL engine for Snowflake Copilot. The data lake vendor claims this fusion archives an execution accuracy rate of 46.4%, outperforming industry leaders including GPT-4 (46%) and CodeLlama 34B Instruct (30.8%).

Snowflake Copilot features and limitations In addition to the text-to-SQL feature, Snowflake Copilot also offers the following features and use cases:

  • Data exploration: Copilot allows users to inquire about their data's structure, refine their analysis with follow-up questions and gain deeper insights, without writing complex queries.
  • Intelligent corrections: Beyond generating queries, Copilot helps users to write cleaner and more efficient SQL queries by suggesting query optimizations, providing explanations and recommending fixes for potential issues in existing queries, streamlining the data analysis process.
  • Snowflake documentation: Copilot can answer queries about Snowflake documentation, for example, it can help look for a specific function or existing features and capabilities.
Snowflake also listed some limitations of Copilot for now, including that the tool does not have access to the data in users' tables; it does not currently support cross-database queries; it can take some time to complete a response and might take up to an hour to recognize newly created databases, schemas, and tables; it may sometimes suggest queries that contain invalid SQL syntax or nonexistent tables or columns.