Generative AI platform
Amazon Bedrock
Serverless access to foundation models and tools for RAG, agents, guardrails, evaluation, customization, prompt management, flows, logging, and throughput.
Key points
- Provides managed API access to foundation models from AWS and third-party model providers.
- Supports text, image, embedding, and multimodal use cases depending on the selected model.
- Keeps infrastructure management away from the builder while still supporting IAM, logging, encryption, and VPC access patterns.
- Includes native capabilities for Knowledge Bases, Agents, Guardrails, model evaluation, prompt management, flows, and model customization.
- Supports on-demand inference and provisioned throughput for workloads that need more predictable capacity.
When to use it
- Choose Bedrock when an application needs foundation model inference without managing model servers.
- Use Bedrock for chat assistants, summarization, content generation, semantic search, and RAG workflows.
- Use Bedrock when governance features such as guardrails, logging, IAM, and private connectivity matter.
Exam tips
- Bedrock is the generative AI service; SageMaker is the broader ML build, train, and deploy platform.
- Knowledge Bases add managed RAG, Agents add task orchestration, and Guardrails add safety controls.
- Customer prompts, responses, and custom model data are not used to train base foundation models.
- Use provisioned throughput for predictable Bedrock inference capacity, not for general SageMaker endpoint scaling.