What Snowflake Cortex AI does
Snowflake Cortex AI adds model functions, document parsing, semantic search, agents, generated SQL, and conversational analytics to the Snowflake data platform.
Snowflake Cortex AI is a family of AI capabilities embedded in the Snowflake platform rather than a standalone warehouse. AI Functions transform structured and unstructured columns, AI Parse Document extracts document content, Cortex Search builds semantic retrieval services, and Agents combine retrieval and structured analytics. Cortex Analyst and Snowflake Intelligence can translate questions into SQL-backed answers using governed semantic definitions and approved tools.
Pricing combines several meters. Supported AI services consume AI Credits by model tokens, indexed data, pages, messages, or feature-specific units, while regional versus global routing changes the AI Credit price. Virtual warehouses, storage, transfers, search serving, generated SQL, orchestration, and non-AI services can add Platform Credits. Forecast both input and output tokens, repeated agent tool calls, search index size, refreshes, query scans, warehouse auto-suspend, and contract discounts from the current consumption table.
Keeping processing near warehouse data does not make every prompt or result safe. Broad roles can expose columns through AI, semantic models can encode wrong joins or metrics, retrieved documents can contain hostile instructions, and generated SQL can scan huge tables, leak small cohorts, mutate objects, or produce plausible but incorrect answers. Enforce least privilege, masking and row policies, restrict models and regions, review semantic definitions, dry-run read-only SQL, cap resources, test against known answers, monitor usage and lineage, and require a data owner to approve production actions.
How Snowflake Cortex AI works
Users invoke Cortex AI Functions from SQL or APIs to classify, extract, embed, summarize, translate, parse, or generate over authorized data. Cortex Search indexes selected content for retrieval, while Cortex Agents can orchestrate Search, Analyst, models, and tools. Cortex Analyst uses semantic models to draft SQL, which executes with Snowflake warehouse compute and the permissions of the configured service context.
How to set up Snowflake Cortex AI
Inventory data, decisions, and risk
Define sources, owners, classifications, residency, roles, semantic metrics, quality contracts, approved models, expected queries, latency, and acceptable AI and warehouse spend.
Create a least-privilege sandbox
Use separate roles, schemas, warehouses, resource monitors, masking and row policies; disable cross-region routing unless approved and expose only representative nonproduction data.
Configure one bounded capability
Pilot an AI Function, Search service, or Analyst semantic model with explicit inputs, descriptions, joins, verified metrics, filters, citations, and safe no-answer behavior.
Evaluate code, answers, and cost
Compare generated SQL and outputs with golden cases, inspect query profiles and bytes, test permission boundaries, injection, nulls, drift, contradictory data, and token amplification.
Promote under human governance
Version models and prompts, require code review and tests, monitor AI and Platform Credits, audit access and lineage, assign incident owners, and revalidate after data or model changes.
Snowflake Cortex AI FAQs
How much does Snowflake Cortex AI cost?
Most current AI features use usage-based AI Credits, while warehouses, storage, transfer, query execution, and other services use Platform Credits. Feature, model, routing, and contract affect price.
Does Cortex AI move data outside Snowflake?
Processing and routing depend on the feature, model, account region, cross-region setting, and terms. Administrators should approve residency and model paths before sensitive use.
Can Cortex Analyst generate SQL?
Yes. It uses semantic definitions to produce SQL-backed answers, but teams must inspect joins, filters, metrics, permissions, query cost, and results before relying on them.
Do Snowflake permissions automatically make an agent safe?
No. Least-privilege roles help, but tool configuration, semantic models, row and masking policies, retrieved instructions, exports, and downstream actions still require testing.
How should teams control Cortex costs?
Track AI usage views and warehouse query history, set budgets and resource monitors, cap inputs and agent steps, suspend unused compute, and forecast index refresh and serving costs.
Listing reviewed 2026-07-15. Product details and pricing can change; verify important terms on the provider's website.
Related Data engineering AI tools
Related AI guides
Reviews
Tell the community what you made, what worked, and what you wish you knew before starting.