Unlocking New Capabilities in Cortex Analyst: Multi-turn Conversations, JOINs, and More!

Cortex Analyst is a fully managed service providing a conversational interface for interacting with structured data in Snowflake, streamlining the development of intuitive, self-serve analytics applications for business users with industry-leading accuracy. Since launching the Cortex Analyst Public Preview in August, we’ve introduced significant advancements to support more complex use cases while maintaining high accuracy. This blog covers the exciting new updates in Cortex Analyst.

“We’ve built and incorporated a Cortex Analyst powered assistant into our data portal to massively simplify interaction with analytical data products in the Cimpress Data Mesh. We anticipate a significant increase in data accessibility and autonomy, the minimization of query errors and lower training costs as we roll it out globally.” ~Michael Theodoulou, Principal Product Manager, Cimpress

Cortex Analyst — Announcements Round up

Joins for Star and Snowflake Schemas now in Public Preview
We’re thrilled to announce that Cortex Analyst now supports SQL joins, enabling advanced data analysis across multiple tables, particularly within star and snowflake schema configurations. This enhancement allows users to query fact and dimension tables with ease, bringing a new depth to data exploration.

Setting up joins is as simple as defining relationships in the YAML configuration file, specifying tables, columns, join types, and relationship types, all while ensuring data integrity with primary key definitions. Optimized for simple star and snowflake schemas, this update empowers users to uncover richer insights directly within Cortex Analyst, expanding its capabilities to deliver a seamless, self-serve analytics experience.

Supporting joins in a text-to-SQL system is far from as simple as just joining two tables together. It requires thoughtful handling of join paths, join types, disambiguation, aggregation, and distinctness, along with careful parsing to generate SQL that accurately reflects the user’s question. Learn more on how we’re approaching the problem in our engineering blog: Cortex Analyst: Introducing Joins for Star and Snowflake Schemas.

Multi-Turn Conversations now in Public Preview
Introducing multi-turn conversations in Cortex Analyst! This feature allows users to ask follow-up questions that build on previous queries, creating a fluid, interactive data exploration experience. Cortex Analyst remembers the context from prior queries, enabling users to refine or expand their questions without restating details, making it easier to dig deeper into data insights.. For example, a user might ask for revenue growth in one region, then follow up with, “What about North America?” Cortex Analyst understands and responds in context, making data analysis even more conversational and user-friendly. Explore our approach in the engineering blog: Cortex Analyst Support for Multi-Turn Conversations.

Cortex Search Integration Public Preview
The recent integration of Cortex Search Service with Cortex Analyst enhances SQL queries generated by Cortex Analyst by enabling dynamic retrieval of exact or similar values for complex, high-cardinality data fields. This integration is especially valuable for dimensions with rapidly changing or high-cardinality values, as it allows Cortex Analyst to access relevant strings directly from dimensions associated with the Cortex Search service. By reducing the need to hard-code sample values in the semantic model, this feature minimizes data duplication, keeps the semantic model lean, and improves SQL query accuracy generated by Cortex Analyst. Dive deeper into this feature on our engineering blog Cortex Analyst and Cortex Search: Better Together.

Introducing API Level Role-Based Access Control (RBAC)
To further enhance security and access management, we’re introducing API-level Role-Based Access Control (RBAC) for Cortex Analyst. Now, the active role in an API request must include the SNOWFLAKE.CORTEX_USER role, granted by default to PUBLIC. For more targeted access, admins can revoke CORTEX_USER from PUBLIC and assign it to specific roles, providing precise access control as needed.

Suggested Questions Public Preview
The new Suggested Questions feature offers users context-aware prompts during data exploration. This feature dynamically adapts based on the semantic model configuration: using LLM-generated questions when no verified queries (VQR) are defined, leveraging VQR-based suggestions when available, and enabling specific queries to be tagged as onboarding questions for tailored guidance. This flexibility allows data teams to configure more effective onboarding flows, giving business users seamless access to meaningful questions and enhancing their experience with Cortex Analyst powered applications.

Enhanced Semantic Model Generation Tooling: Now with Third-Party Semantic Layer Translation Support

We’ve enhanced the Semantic Model Generator for Cortex Analyst to simplify Semantic Model YAML file creation. Available both as a local Streamlit app and a Streamlit In Snowflake (SiS) app, the Generator uses a Streamlit front end to analyze selected tables, curating metadata for columns, data types, and descriptions. It now also includes a Partner Translator module, which translates third-party semantic models and merges relevant partner metadata with metadata into the Cortex Analyst Semantic File format. As of today, the module supports translation from dbt semantic models and Looker Explores that reference Snowflake views and tables.

Snowsight UI Semantic Model Admin Tooling (Private Preview coming soon)
The upcoming private preview of the Snowsight UI for Semantic Model Administration is a fully hosted, intuitive interface centralizing semantic model creation, management, and iteration. This interface will also support model usage monitoring and provide smart suggestions based on business user feedback, making model management more responsive and impactful. Stay tuned for updates!

Customer Success Stories: Empowering Data-Driven Decisions with Cortex Analyst

Since its recent launch, Cortex Analyst has gained significant traction, empowering hundreds of customers to drive data accessibility across their organizations. Major companies like Bayer and JPMC are enhancing their business intelligence efforts by adopting Cortex Analyst’s self-serve, natural language query capabilities. End users — from executives to sales teams and even external customers — are leveraging this tool to access faster, more insightful data-driven answers, transforming how they interact with and act upon business data.

Conclusion

Below are some useful resources to get you started and learn more about Cortex Analyst. We can’t wait to see the amazing ways you’ll use Cortex Analyst. We’re just getting started, so stay tuned — there’s a lot more exciting stuff coming your way!

• Be sure to register for BUILD and tune in to the opening keynote and the Paving the Way for Reliable Self-Serve Analytics with Cortex Analyst session to catch all these announcements in action!
• Build your first Cortex Analyst-powered chat app using this quickstart guide
• Follow our quick start for using our Semantic Model Generator Tooling for Creating Semantic Models for Cortex Analyst
• Learn how you can Translate Existing 3rd party Semantic Layers for Use by Cortex Analyst
• Checkout our Github Cortex Analyst Samples repository for inspiration

For further details on the feature, as well as best practices for obtaining more accurate results, be sure to check out the Cortex Analyst documentation.