//
Search across all documentation pages
9 pages in this section.
The client-messages-response mental model behind Amazon Bedrock's Converse API, and how one model-agnostic surface reaches Claude, Llama, Nova, and Titan without per-model JSON.
Your first Converse calls against a Bedrock foundation model in boto3 and AWS SDK v3 - system prompts, inference config, multi-turn chat, token usage, listing models, and a first stream.
The two Bedrock Runtime inference surfaces compared - model-agnostic Converse vs model-specific InvokeModel - with real request bodies in boto3 and AWS SDK v3, and when each one wins.
Stream Bedrock model output token by token with ConverseStream in boto3 and AWS SDK v3 - the event lifecycle, assembling text, reading the final usage event, and error handling.
The two Bedrock capacity models compared - pay-per-token on-demand vs reserved Provisioned Throughput - with the CreateProvisionedModelThroughput control-plane calls in boto3 and AWS SDK v3.
A checklist for running Amazon Bedrock inference from the SDK safely and cheaply - API choice, prompt/response handling, retries and reliability, token-cost monitoring, and IAM security.