Skip to main content
This lesson continues preparing for examples that call Amazon Bedrock foundation models, showing common usage patterns, request flow, privacy considerations, and extension points. Read on to understand how Bedrock hosts and serves models so you can design integrations that are simple, scalable, and vendor-agnostic.
In Bedrock, the term “foundation model” refers to a vendor-hosted model listed in the Bedrock catalog. Practitioners often shorten this to “model” — both terms mean the same hosted model instance available through Bedrock.
Common usage patterns for Bedrock foundation models
  • Text generation: send text prompts to Bedrock and receive generated text (examples: summarization, chatbots, Q&A).
  • Multimodal requests: submit text plus images (examples: image Q&A, product-photo categorization). The model typically returns a textual response for classification or description.
  • Image generation: supply a text prompt and receive a generated image file (e.g., JPEG/PNG) for marketing and creative content.
  • Image editing: send an existing image plus an instruction prompt, and receive the modified image (examples: object removal, inpainting).
A presentation slide titled "Solution: Bedrock Foundation Models" showing a four-column table of model types (Text Generation, Multimodal, Image Generation, Image Editing), what you send and receive via the API, and typical use cases. The table outlines inputs (text, text+image) and outputs (generated text, text response, generated/modified images) for each model.
Bedrock is a fully managed AWS service Bedrock removes the need to provision or maintain servers, handle availability zones, or manage scaling and failover. AWS operates the underlying compute and availability infrastructure so you can focus on integrating models into your application. For comparison, Bedrock follows the same managed-service model as other AWS services like Amazon S3 (object storage), Amazon SQS (message queuing), and Amazon DynamoDB (NoSQL). See the vendor docs for each service to understand the managed responsibilities and SLAs:
A slide titled "Solution: AWS Fully Managed Services" showing icons for S3, SQS, and DynamoDB inside a dashed box. Below the services is a "Service User" icon representing a user interacting with those managed services.
Because Bedrock is fully managed:
  • You do not provision servers, configure availability zones, subnets, or manage failover.
  • AWS handles capacity management, scaling, and high availability.
  • To get started you only need to call the Bedrock Runtime API from your application, specifying a prompt and a model name.
Model providers and vendor choice Bedrock hosts foundation models from multiple vendors. Common providers in the Bedrock catalog include Amazon, Anthropic, Cohere, DeepSeek, Meta, Mistral, NVIDIA, and Stability AI — the catalog continues to grow. You pick the vendor and model that best match your performance, cost, and capability needs.
A presentation slide titled "Solution: Bedrock Model Providers." It shows logos of model providers including Amazon, Anthropic, Cohere, Deepseek, Meta, Mistral AI, NVIDIA and stability.ai.
Because Bedrock exposes a single Runtime API, switching vendors or mixing models in an application requires minimal change: update the model name parameter in your requests. Choosing a model (practical tips)
  • Browse the Bedrock model catalog for the AWS Region you operate in — availability is region-specific.
  • Check each model’s supported modalities (text, image, multimodal) and performance characteristics.
  • Models listed in the catalog are pre-deployed on Bedrock’s managed compute; you will not manage the underlying infrastructure directly.
A presentation slide titled "Workflow: How to Use Bedrock?" showing a catalog of eight GenAI model tiles on the left. On the right are three green checkmarks with benefits: "Deployed on managed compute platform," "No servers to manage," and "No infrastructure responsibility."
Request flow overview
  1. End user interacts with your application (for example, a web or mobile UI).
  2. Your application server uses an AWS SDK or the Bedrock Runtime API to call Bedrock. (Clients should not call Bedrock directly.)
  3. The application supplies two key parameters:
    • The prompt or input (text, image, or both).
    • The Bedrock model name to route the request.
  4. Bedrock routes the prompt to the selected hosted model (for example, meta-llama/llama-3-1-instruct-8b).
  5. The model generates the response; Bedrock returns that response to your application, which decides how to present or use it.
This pattern keeps your integration consistent across different models: just change the model name to route requests to a different foundation model. Example (pseudocode)
Privacy and data handling
  • Data you want processed (for example, documents to be summarized) must be included in the prompt you send to Bedrock.
  • Your contractual relationship is with AWS as the managed host; you are not required to enter separate commercial agreements with the model vendors to use models via Bedrock.
  • AWS provides controls and documentation about data handling, retention, and whether data is used for model training. Review the official documentation and your AWS agreement for details.
Before sending sensitive or regulated data to Bedrock, confirm your organization’s compliance requirements and review the Bedrock data protection and retention policies: https://docs.aws.amazon.com/bedrock/latest/userguide/data-protection.html
A dark presentation slide titled "Results" with a circular diagram of four colored icons. Each label lists benefits: no commercial relationship required, data remains within AWS (ensuring privacy), data is not used to train provider models, and a single API enables access to any model.
Extensibility and external tooling Bedrock can be augmented with additional tooling — APIs, external databases, knowledge bases, or tool integrations — so that models can participate in richer workflows (for example, retrieval-augmented generation, tool invocation, or external API calls). This allows you to build systems beyond single prompt/response interactions. Key takeaway Bedrock is a fully managed AWS service that provides vendor-agnostic access to multiple foundation models through a single API. AWS manages the infrastructure, scaling, and operational responsibility so you can focus on integrating the right model(s) for your application.
A presentation slide titled "Key Takeaway" that states: "Amazon Bedrock is a fully managed AWS service that provides access to multiple foundation models through a single API."
Next up: the following lesson will explain the basic Bedrock architecture and show concrete code examples for calling Bedrock models from popular SDKs.

Watch Video