Aws Bedrock

ThrottlingException for concurrent request limits (model-specific; Claude has generous limits). ModelStreamErrorException for stream interruptions. ValidationException for invalid request format (check API format matches model provider spec). Bedrock SLA: 99.9%. Model availability varies by region — check which models are available in target region before deploying. On-demand pricing per token.

docs.aws.amazon.com/bedrock covers model provider API formats, Knowledge Bases, Agents, prompt management, and guardrails. Getting started: enable model access in AWS console, install boto3, first invoke in under 5 minutes. Model API formats documented per-provider (Anthropic Messages API, Meta Llama Chat API, etc.). Community via AWS Forums and re:Invent.

Bedrock provides a single AWS-native API to access foundation models from multiple providers — Anthropic Claude, Meta Llama, Amazon Titan, Mistral, Cohere, AI21, and Stability AI — without managing model infrastructure. For agents on AWS: use bedrock-runtime to invoke models for text generation, embedding, and image generation; Bedrock Knowledge Bases for managed RAG; Bedrock Agents for agentic workflows with tool definitions. Tight IAM integration makes it straightforward for AWS-hosted agent systems. Confidence is docs-derived.

AWS SDK (boto3): br = boto3.client("bedrock-runtime", region_name="us-east-1"). response = br.invoke_model(modelId="anthropic.claude-3-5-sonnet-20241022-v2:0", body=json.dumps({"anthropic_version": "bedrock-2023-05-31", "messages": [{"role": "user", "content": prompt}], "max_tokens": 1000})). response_body = json.loads(response["body"].read()). Streaming: br.invoke_model_with_response_stream(...). Knowledge Bases RAG: bedrock-agent-runtime client, retrieve_and_generate. Bedrock Agents: create agents with tool definitions via bedrock-agent management client.

IAM auth: AWS credentials. Per-operation: bedrock:InvokeModel, bedrock:InvokeModelWithResponseStream per model ARN. Bedrock Knowledge Bases: bedrock:RetrieveAndGenerate. IAM role for Lambda/ECS access. Managed identity (EC2/Lambda). HTTPS enforced. Cross-region inference profiles available.