What happened
Google Cloud announced that API Gateway now supports streaming in Public Preview. This update allows developers to route real-time, long-lived connections—such as those used by AI models generating tokens or agents reporting progress—through the same gateway infrastructure used for standard request-response APIs. The supports streams open for up to one hour and integrates with model routing to handle responses from Gemini, Claude, and OpenAI gpt-oss models. Additionally, the update introduces deterministic default hostnames, allowing developers to predict and manage gateway URLs in infrastructure-as-code environments before deployment.
Google Cloud has introduced streaming support to API Gateway, currently available in Public Preview. This allows developers to manage real-time, bidirectional data flows through the same gateway infrastructure used for traditional request-response APIs. Previously, developers had to choose between using API Gateway for governance and losing streaming capabilities, or bypassing the gateway for streaming endpoints and losing authentication and observability controls.
The update supports the full range of modern streaming patterns, including Server-Sent Events (SSE) and WebSockets, with streams remaining open for up to one hour. The implementation requires no changes to existing API specifications; the same document that describes standard APIs now describes streaming operations. During the Public Preview phase, developers must opt in by setting a specific flag when creating the gateway. At General Availability, streaming is planned to be the default behavior without requiring opt-in.
The announcement highlights the integration of streaming with model routing capabilities. API Gateway can now expose a single OpenAI-compatible endpoint that dispatches requests to Gemini, Claude, or OpenAI gpt-oss models based on the request payload. This allows clients to receive streamed responses from these models while maintaining centralized control over model access, authentication, and rate limiting.
A secondary update introduces deterministic default hostnames for gateways. Previously, default hostnames included a generated hash, making URLs unpredictable until after resource creation. Now, the hostname is a pure function of user-chosen values, allowing developers to generate, template, and commit URLs to source control before the gateway exists. This change simplifies infrastructure-as-code workflows and environment promotion.
Source details: developers.googleblog.com ↗
Why it matters
This update addresses a critical gap in API management for AI applications, where streaming is the expected interaction pattern rather than an exception. By supporting streaming, API Gateway allows organizations to maintain centralized governance, authentication, and observability for AI traffic, preventing the fragmentation of security policies between standard and real-time endpoints. This is particularly important for agentic workflows that rely on long-lived connections for tool-calling and intermediate step reporting. The deterministic hostname further reduces operational friction for teams managing environments as code, ensuring consistent configuration across development, staging, and production cycles.
Streaming is the standard interaction pattern for AI applications, where users expect to see model responses appear as they are generated. Agents also rely on streaming to report intermediate steps and manage tool-calling protocols. An API management layer that cannot stream limits the user experience and operational efficiency of AI systems.
By bringing streaming into API Gateway, Google Cloud enables organizations to apply consistent security policies, authentication, and observability to AI traffic. This prevents the creation of ungoverned 'shadow' endpoints that bypass standard security controls, which is a significant risk in environments where AI endpoints consume expensive quotas and require strict access control.
The deterministic hostname addresses a common operational friction point in DevOps. By allowing developers to know the gateway URL before deployment, it removes bootstrapping problems in CI/CD pipelines, firewall configurations, and partner integrations. This is particularly useful for teams managing multiple environments or promoting configurations from development to production.
This update positions API Gateway as a lightweight entry point to Google Cloud's broader AI gateway capabilities, which include Apigee and Agent Gateway. It allows teams to choose the appropriate level of governance for their specific use cases, from simple model routing to complex agent connectivity management.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
What to watch next
Monitor the transition from Public Preview to General Availability (GA), where streaming is planned to become the default behavior without opt-in flags. Observe how organizations adopt the combined use of API Gateway, Apigee, and Agent Gateway to manage complex agentic traffic. Watch for documentation updates regarding best practices for reconnection logic in long-lived streams and the specific policy controls available for securing AI endpoints.
The transition from Public Preview to General Availability is a key milestone. At GA, streaming is expected to become the default behavior, removing the need for opt-in flags. This will simplify the developer experience and standardize the behavior of API Gateway across all environments.
Adoption of the combined gateway strategy, where API Gateway, Apigee, and Agent Gateway are used together, will be important to monitor. This composable approach allows for granular control over different types of traffic, such as routing agent egress through Agent Gateway for security and then passing requests to API Gateway for model routing.
Documentation and best practices for handling long-lived connections will be critical. Since streams can stay open for up to one hour, developers will need to implement robust reconnection logic to ensure session continuity. Google's guidance on this will be essential for reliable production deployments.
The impact on existing API management workflows will be significant. Teams will need to update their infrastructure-as-code templates and CI/CD pipelines to leverage the deterministic hostname and the new streaming capabilities. This may require changes to how environments are provisioned and managed.