Snowflake has launched Cortex Agents, a fully managed service aimed at integrating, retrieving, and processing structured and unstructured data at scale.
Cortex Agents plan tasks, execute them using tools, and refine responses for improved accuracy. Available via a REST API, the service integrates into applications using Cortex Analyst for structured data (SQL generation) and Cortex Search for unstructured data retrieval.
The new solution, now in public preview, enables businesses to build AI-driven applications with advanced governance and security features.
The Cortex Agents framework integrates Anthropic’s Claude 3.5 Sonnet, which enhances reasoning, coding, and workflow execution within Snowflake’s secure perimeter.
Cortex Agents review user queries and break them into structured and unstructured components. For example, a business user can request top distributors by revenue (structured) and then ask about contract details (unstructured). The agent disambiguates queries, splits tasks, and selects tools for execution.
Cortex Analyst, a fully managed LLM-powered Snowflake Cortex feature, generates SQL for structured data, while Cortex Search retrieves insights from text, audio, and images. The system ensures governed access and compliance while mapping business terms to structured data using semantic understanding.
Snowflake reports that Cortex Analyst has achieved 90% accuracy in text-to-SQL use cases.
On the other hand, Cortex Search has outperformed OpenAI’s embedding models by 12% in unstructured data retrieval accuracy. It supports large-scale indexing, improved affordability, and customisable vector embeddings.
Snowflake also introduced Cortex AI Observability, powered by TruLens, for the evaluation and tracing of AI agents. “AI observability can evaluate agent performance using techniques such as LLM-as-a-judge, allowing customers to refine and optimise their applications,” Snowflake said.
Snowflake sees AI agents as a transformative force for enterprises, automating complex tasks and improving efficiency across industries such as finance, engineering, and customer support. “As LLMs continue to advance, agents will collaborate, plan, execute, and refine tasks, driving efficiency and reducing costs,” the company said.
