# Google Cloud's API Gateway adds streaming support in public preview

published: 2026-10-22T15:05:41.777Z
topic: infrastructure
source: https://developers.googleblog.com/streaming-comes-to-api-gateway-real-time-and-ai-ready-apis-without-changing-your-spec/
url: https://aiagentnews.io/stories/google-cloud-s-api-gateway-adds-streaming-support-in-public-preview

Google Cloud has added streaming support to API Gateway in Public Preview, so Server-Sent Events, WebSocket and gRPC streaming endpoints can sit behind the same gateway policies as other APIs. The feature is turned on with a flag at gateway creation, and new gateways also get a predictable default hostname.

Until now, developers who wanted streaming through API Gateway had to either give up streaming or send streaming traffic around the gateway, losing authentication, API key checks and observability. The preview covers Server-Sent Events, chunked transfer and NDJSON responses, WebSockets, and gRPC server-side and bidirectional streaming. Streams can stay open for up to one hour, so clients should reconnect when a stream ends.

The API contract does not change, and no streaming-specific extensions or separate spec are needed. API key enforcement, JWT validation, service account authentication to backends, Cloud Logging and Cloud Monitoring, and path routing keep working. The feature is enabled with a flag when a gateway is created, and existing gateways cannot be converted in place. Gateway hostnames are now built from the gateway ID, project number, region and a fixed domain, rather than a generated hash.

For AI workloads, model routing can stream responses from Gemini, Claude and OpenAI gpt-oss models in Model Garden in Gemini platform. Latency metrics now measure the length of a stream rather than time to first byte. Custom domains, Console enablement and Terraform support are due to arrive during the preview.

## Why it matters if you build agents

Agent builders can expose streamed model output and intermediate agent steps through the same gateway policies used for their other APIs, instead of running a separate proxy for streaming routes. Clients still need reconnection logic, since streams can last up to an hour. Latency dashboards built on request duration will show stream length, so they need to be updated before reading them as response times.