Cortex Agents

What it does

Cortex Agents takes a natural-language request from a user or application and returns a generated response. An agent plans the work, queries structured data through semantic views, retrieves from unstructured sources, can run Python in an isolated sandbox when the code execution tool is enabled, and reasons over the results across several steps. Threads keep conversation context across turns.

Teams use it to answer questions that combine tables and documents inside Snowflake's governed environment. Data access follows existing Snowflake roles and privileges, and Snowflake manages the orchestration runtime rather than requiring you to build your own loop or sandbox. Agents can be created in Snowsight, with SQL, or through the REST API, then called from your own application or used in Snowflake CoWork and Snowflake CoCo.

Cortex Agents suits data and engineering teams already working in Snowflake. Calling an agent requires the SNOWFLAKE.CORTEX_USER or SNOWFLAKE.CORTEX_AGENT_USER database role and privileges on the agent and its tools. Responses are not guaranteed to be accurate, so answers should be reviewed before reaching end users. Streamlit in Snowflake apps on a warehouse runtime cannot call the agent APIs. Agent orchestration is billed in AI Credits per token, while tools such as custom tools and warehouses add their own costs.

What it can do

• Generates SQL over structured data using Cortex Analyst semantic views
• Retrieves information from unstructured sources with Cortex Search
• Runs Python in a secure, isolated sandbox for calculations when the code execution tool is enabled
• Calls custom tools built from stored procedures and UDFs
• Connects to remote MCP servers such as Atlassian Jira and Salesforce
• Maintains conversation state across turns with threads

Works with

• Cortex Analyst
• Cortex Search
• Atlassian Jira (via MCP connector)
• Salesforce (via MCP connector)

Before you choose

• Requires SNOWFLAKE.CORTEX_USER or SNOWFLAKE.CORTEX_AGENT_USER role plus privileges on the agent and tool objects.
• Agent orchestration is billed per token in AI Credits; tools such as custom tools (warehouse compute), Cortex Search, and web search or MCP connectors add separate costs, and not all underlying services use AI Credits.
• Responses and citations are not guaranteed accurate; review answers before serving users.

Listed under

• Coding
• MCP and agent tools

Get started

Start from the Get started guide for a first agent.

1. Create an agent in Snowsight, with SQL, or through the REST API, including name, model, and instructions.
2. Add tools such as semantic views, search services, and warehouses the agent needs.
3. Test the agent in the Snowsight playground, then integrate it via the agent:run REST API using threads.

Orchestration is billed in AI Credits per million tokens, and the rate depends on the orchestration model; see the Snowflake Service Consumption Table. Costs add up across the tools an agent calls, but not all underlying services use AI Credits: warehouse compute for custom tools and SQL execution is billed at standard virtual warehouse rates, while services such as Cortex Search and Cortex Analyst have their own usage meters.

Source: docs.snowflake.com/en/user-guide/snowflake-cortex/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Cortex Agents | Snowflake Documentation ↗docs.snowflake.com
• Snowflake AI pricing | Snowflake Documentation ↗docs.snowflake.com

Company website

docs.snowflake.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.