Snowflake makes AI Gateway core features generally available; trace analytics stays preview
Snowflake’s November 3 release notes make inference, observability and cost management for Cortex AI Gateway generally available. The gateway gives an organization a governed endpoint for model requests. Not every related feature has left preview.
For inference, the gateway accepts requests in OpenAI Chat Completions or Anthropic Messages API formats. Snowflake says clients can use the official SDKs or a coding agent to send requests to that endpoint.
For observability, AGENT_TRACE_TABLE lets teams query request spans, including the models called, timing, token counts and errors. Client applications can also export their own OpenTelemetry traces to the gateway.
Cost-management features include the AI_GATEWAY_USAGE_HISTORY view for spend reporting, shared resource budgets for team-level tracking and per-user quotas that can block an individual user at a limit. Those are distinct controls; spend reporting alone is not a spending cap.
Trace analytics is newly available in preview. In Snowsight, the Analytics tab on the gateway’s Observability page can chart model usage, request failures, latency, MCP tool calls and conversation length from the trace table without SQL.
Tools for Cortex AI Gateway also remains in preview. Teams can start from AI Gateway in Snowsight, while checking the separate status and setup requirements of each capability. This release note does not establish that all gateway features are generally available or free.