AgentCore Gateway now supports Bedrock InvokeModel APIs and model fallback

Amazon Bedrock AgentCore Gateway now accepts Amazon Bedrock InvokeModel and InvokeModelWithResponseStream API calls on its inference endpoint, allowing existing SDK integrations to reach any model provider through a single governed endpoint without client-side code changes. This unified entry point removes the need to rewrite applications for each provider, simplifying deployment while maintaining governance controls. The change applies to all AWS Regions where AgentCore Gateway is available, excluding those in China.

Model fallback routing automatically redirects requests when the primary destination is throttled, out of capacity, or unreachable, keeping agents operational without clients implementing custom retry logic. Administrators define an ordered chain of alternative destinations in a gateway rule and specify conditions such as throttling, server errors, or response timeouts to trigger a switch. The system skips failing destinations via a built-in circuit breaker, reducing latency and avoiding repeated failed attempts.

Fallback routing operates transparently, so clients send a single request and receive one response even when the gateway internally tests multiple destinations. This behind-the-scenes process ensures consistent behavior for applications while improving resilience and uptime. The feature is available in all supported AWS Regions except those in China.

To implement these capabilities, users configure gateway rules in the Amazon Bedrock AgentCore Gateway User Guide, which outlines how to set up ordered fallback chains and circuit breaker thresholds. The guide provides step-by-step instructions for integrating InvokeModel APIs and enabling model fallback within existing agent workflows.