Sep 15, 2026: Cortex AI Gateway (Preview)

We are pleased to announce the preview of Cortex AI Gateway, the control plane where you govern how AI applications and third-party agents in your organization reach models.

It gives platform, security, and FinOps teams a single place to grant model access, attribute spend, and audit AI activity, while application teams get a standard endpoint to build against.

Snowflake provisions the gateway for you. Each account has a single gateway object named SNOWFLAKE. ACCOUNTADMIN can grant and revoke privileges on it for different roles. USAGE is granted to PUBLIC by default, so the gateway is reachable out of the box; revoke it if you want to restrict traffic to specific roles.

With this preview, you can:

  • Send inference through the gateway endpoint using the OpenAI Chat Completions or Anthropic Messages API formats, from the official SDKs or a coding agent such as OpenCode. For details, see Inference with Cortex AI Gateway.
  • Inspect what agents did with a per-gateway trace table. Query spans to see the models called, step timing, token counts, and errors for any request. For details, see Observability for Cortex AI Gateway.
  • Report on usage with the new AI_GATEWAY_USAGE_HISTORY view in the ACCOUNT_USAGE schema, which records the gateway, user, model, and token consumption for each request. Snowsight also surfaces agent activity out of the box, with request and response counts, token counts, and latency across models, so you can inspect what agents are doing without writing SQL.
  • Control spend by adding the gateway as a shared resource to a custom budget, so you can track credit consumption by the team or cost center consuming it. Per-user quotas block individual users at a limit. For details, see Using budgets for AI features (shared resources) and Per-user quotas. For guidance on choosing between them, see Cost management for Cortex AI Gateway. Snowsight charts gateway credit usage across models and users, so you can see where spend is going before setting a limit.

Inference through the gateway respects model access control: a request still needs whatever access the underlying model requires, such as the SNOWFLAKE.CORTEX_USER database role or the controls described in Privileges and model access for Cortex AI Functions. The gateway governs the path traffic takes, not the set of models a user is entitled to, so it doesn’t expand your security perimeter. Traffic is recorded at the metadata level by default, with prompts and responses captured only if you turn payload logging on.

To get started, sign in to Snowsight and select AI & ML » Cortex AI Gateway.

For more information, see Cortex AI Gateway.