Question

A developer builds a customer-facing chatbot where a Lambda function calls Bedrock's InvokeModelWithResponseStream API and must deliver tokens to the browser as they are generated. The Lambda is fronted by a REST API Gateway endpoint. During testing, the browser receives nothing until the model has fully finished generating, after which the entire response arrives at once. The team needs true incremental token delivery to the client without a major re-architecture of the compute layer. Which change best achieves streaming behavior end-to-end?

Accepted Answer

Switch the Lambda to use a Lambda response streaming enabled Function URL (or invoke mode RESPONSE_STREAM) and have the client consume the streamed HTTP response directly, bypassing the buffering REST API Gateway integration.

Answer

Increase the Lambda memory allocation and timeout so the function can flush partial results more frequently through the existing REST API Gateway integration.

Answer

Enable payload compression and chunked transfer encoding on the REST API Gateway stage to force incremental delivery of the buffered Lambda response.

Answer

Replace InvokeModelWithResponseStream with the synchronous InvokeModel API and poll the response in a client-side loop to simulate streaming.