🌊Google Cloud API Gateway Adds Streaming for AI Apps With 1-Hour Connections

Your AI streaming endpoints can finally sit behind a gateway

Google Cloud API Gateway can now sit in front of streaming endpoints, and that's the part that matters for AI teams. You get SSE, chunked transfer, HTTP response streaming, WebSockets, and gRPC without giving up the policies you already rely on. If you're shipping chat UIs, agent loops, or anything that streams tokens, this closes a gap. Until now, putting a gateway in front of that traffic meant either bypassing your controls or rewriting specs. Here, API keys, JWT validation, service account auth to backends, and path routing all keep working. Pair it with model routing and you get one OpenAI-compatible endpoint that sends requests to Gemini, Claude, or gpt-oss based on the model field, with central control over who reaches what. The catch? Streams max out at one hour, so clients need to reconnect for longer sessions. Gateway hostnames are now predictable too, following GATEWAY_ID-PROJECT_NUMBER.REGION.gateway.dev, instead of a hash you only learned after creation. During preview, you have to pass --enable-streaming when creating a gateway. Worth watching: how this behaves at scale once the preview flag comes off.