Cortex AI Gateway

What it does

Cortex AI Gateway is an open preview Snowflake feature for governing how AI clients in an organization reach models. It takes model inference requests from coding agents, third-party clients, and SDKs, sends them through a single endpoint, and returns model responses. Each account has one gateway object, named SNOWFLAKE, that is created automatically.

Platform teams use it to control which roles can send requests, attribute gateway spend to users and teams, and review traces covering models called, timing, token counts, and errors. Budgets and per-user quotas can limit spend, and revoking USAGE stops a role from sending requests. Prompt and response capture is optional and off by default.

It suits platform and cost-management teams in accounts in supported AWS, Azure, and GCP commercial regions, excluding AWS Europe (Spain). Access requires Snowsight and appropriate privileges. Model usage is billed in AI Credits based on model and token consumption; see the published Snowflake AI pricing page. Budget evaluation is periodic and can lag by up to 6.5 hours, while per-user quotas block within minutes.

What it can do

• Single governed endpoint for model inference requests from agents, clients, and SDKs
• Traces and spans for each request, including models, timing, tokens, and errors
• Per-team and per-user cost attribution through the AI_GATEWAY_USAGE_HISTORY view
• Budgets and per-user quotas for gateway spend
• Privilege-based access control through USAGE and MONITOR grants
• Optional prompt and response capture with per-user tag exclusions

Before you choose

• Model access is still controlled by the underlying model permissions, such as SNOWFLAKE.CORTEX_USER; gateway USAGE does not widen model entitlement.
• Payload capture records prompts and responses, so treat the trace table as sensitive once enabled.
• Budget evaluation is periodic and can pass a threshold before action; quota blocks apply within minutes.

Listed under

• Security

Get started

Sign in to Snowsight and select AI & ML > Cortex AI Gateway.

1. Sign in to Snowsight and select AI & ML > Cortex AI Gateway; the gateway object named SNOWFLAKE already exists.
2. Run SHOW AI GATEWAYS to get the endpoint, and grant USAGE to roles that should send requests, or MONITOR to cost reviewers.
3. Review the specification with DESCRIBE AI GATEWAY SNOWFLAKE and enable logging or payload capture only if your governance policy allows it.

Model usage is billed in AI Credits based on model and token consumption. The rate per model is listed in Snowflake's published pricing and consumption tables at https://docs.snowflake.com/en/user-guide/snowflake-cortex/pricing. Budget thresholds are evaluated periodically and can lag by up to 6.5 hours (or one hour with the low-latency option), whereas per-user quotas are enforced within minutes and are separate from periodic budgets.

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 AI Gateway | Snowflake Documentation ↗docs.snowflake.com
• Cost management for Cortex AI Gateway | Snowflake Documentation ↗docs.snowflake.com
• Snowflake AI pricing | Snowflake Documentation ↗docs.snowflake.com

Company website

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

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.