Cloud service · AWS · Updated

Amazon Bedrock - Amazon Bedrock is a managed AWS service for building generative AI applications with foundation models through APIs.

In short

Amazon Bedrock provides access to foundation models from Amazon and other model providers, along with capabilities for retrieval, customization, tool use, and safety controls. Engineers use it to build applications that generate or analyze content without operating the underlying model-serving infrastructure.

Cloud Services Hub →

Free Amazon Bedrock tools

Amazon Bedrock Model FinderFind the models available in a Region and the model ID or inference profile ID to call each one with.Open →Amazon Bedrock Throttling CalculatorSee how many requests per minute and at once a model's quotas allow, and why you get a 429 at low traffic.Open →Amazon Bedrock AgentCore Error DecoderPaste an AgentCore Runtime, Gateway or auth error and see its causes, or a JWT and see what the authorizer checks.Open →Amazon Bedrock AgentCore Dockerfile CheckerCheck a Dockerfile against the AgentCore Runtime container contract: ARM64, 0.0.0.0, the protocol's port, USER.Open →Amazon Bedrock AgentCore Limits CheckerDescribe your agent and see the AgentCore limits it will hit and what a Runtime session costs in CPU and memory.Open →Amazon Bedrock AgentCore Gateway Target ConfiguratorPick a Gateway target and see the outbound auth it takes, the credential provider and the service role policy.Open →Amazon Bedrock AgentCore Token Flow PlannerPlan how Claude, Claude Code, VS Code, Cursor or your code gets a token AgentCore Gateway or Runtime accepts.Open →Amazon Bedrock AgentCore IaC Security CheckerCheck AgentCore resources in a CloudFormation template or Terraform for open gateways, secrets and public networks.Open →Amazon Bedrock Billing DecoderDecode Bedrock usage types, Cost Explorer CSVs and CUR rows into spend per model and token type, and split costs by application.Open →Amazon Bedrock Pricing CalculatorPrice a workload on any Bedrock model with Flex, Priority, batch, Global routing, prompt caching and Reserved, cheapest first.Open →Amazon Bedrock Fine-Tuning Cost CalculatorAdd up training, storage and Provisioned Throughput or on-demand serving of a fine-tuned Bedrock model.Open →Amazon Bedrock InvokeModel Body Generator and ValidatorGet the InvokeModel body and Converse request for any Bedrock model, check a body, or fix "Malformed input request".Open →Amazon Bedrock Batch Inference JSONL ValidatorCheck a batch job's JSONL records against the model's body format and quotas, its S3 layout and why a job failed.Open →Amazon Bedrock Knowledge Bases Configurator and ValidatorGet the vector dimensions, index mapping, role policies, hard limits and idle cost of a knowledge base's vector store.Open →AI Coding Tools on Amazon Bedrock ConfiguratorSet up Claude Code, Cline, Cursor or Codex on Bedrock: settings, inference profile ID, IAM policy, quotas and errors.Open →

9 key features of Amazon Bedrock

01
Managed foundation model access

Invoke models from Amazon and other providers without provisioning model-serving clusters. Check the selected model's supported modalities, Region availability, access requirements, and quotas before integrating it.

02
Converse and streaming APIs

Use Converse and ConverseStream for a consistent conversational interface across supported models. Keep model-specific settings and capability checks explicit when switching providers or model versions.

03
Knowledge Bases for retrieval

Build retrieval-augmented generation workflows using supported data sources and retrieval stores. Tune chunking, metadata filters, and retrieval settings, and evaluate whether retrieved passages actually support the generated answer.

04
Agents and tool orchestration

Configure agents to retrieve information and invoke defined application actions. Validate tool arguments and enforce authorization in the action implementation rather than relying on the model to enforce business rules.

05
Configurable guardrails

Apply supported checks for harmful content, denied topics, sensitive information, and grounding. Test both false positives and missed violations, and retain application-level controls for security-critical decisions.

06
Model customization

Customize supported models using available techniques such as fine-tuning or continued pre-training. Supported methods, training requirements, and deployment options vary by model and Region.

07
Evaluation workflows

Use supported evaluation workflows to compare model outputs or assess retrieval and generation quality. Combine automated measurements with representative test cases and human review where correctness is subjective or consequential.

08
Flexible inference options

Choose among supported on-demand, batch, and provisioned inference options. Evaluate expected traffic, latency requirements, and model eligibility before selecting a capacity and billing approach.

09
AWS security and observability

Control API access with IAM and use supported private endpoints and encryption settings. Combine CloudWatch metrics, optional invocation logging, and CloudTrail records to investigate performance and access issues.

Where Amazon Bedrock is used

  1. 01
    Internal knowledge assistants

    Connect an assistant to approved manuals, policies, and operational documentation using retrieval-augmented generation. Enforce document access restrictions and expose source references so users can check answers.

  2. 02
    Customer support assistance

    Generate draft responses, summarize conversations, and retrieve relevant support articles. Require human review or deterministic approval checks before sending sensitive responses or changing customer accounts.

  3. 03
    Document analysis and extraction

    Extract fields, classify documents, or summarize long reports using models that support the required input format. Validate extracted values against schemas and business rules before writing them to downstream systems.

  4. 04
    Tool-enabled business workflows

    Use agents or application-managed tool calls to check inventory, retrieve order status, or prepare service requests. Run actions under narrowly scoped permissions and require confirmation for irreversible changes.

  5. 05
    Batch content processing

    Process eligible workloads such as catalog enrichment, document labeling, or large-scale summarization through batch inference. Use job tracking, output validation, and retry logic to handle partial failures.

How Amazon Bedrock is priced

On-demand inference

Pay for model usage without reserving dedicated capacity. Text models commonly charge separately for input and output tokens, while other modalities may use units such as images or audio duration; rates depend on the model and inference option.

Provisioned Throughput

For eligible models, purchase provisioned model capacity billed over time, with available commitment terms depending on the offering. Compare expected utilization and throughput needs against on-demand inference before committing.

Batch inference

Eligible models support asynchronous batch jobs with pricing separate from interactive inference. Use this option for workloads that can tolerate job completion latency, and account for associated storage costs.

Model customization

Customization can incur training charges, storage charges for customized models, and charges for subsequent inference or required hosting capacity. Billing units and supported deployment options vary by model and customization method.

Additional capability and supporting service charges

Capabilities such as guardrails, evaluation, and some knowledge base processing operations can add charges beyond model inference. Include embedding generation, vector storage, S3, Lambda, logging, and data transfer where applicable in workload cost estimates.

Inference service tiers

Amazon Bedrock offers Standard, Flex, Priority, and Reserved service tiers.

Services Amazon Bedrock works with

A common architecture places an application backend in front of Bedrock and uses an IAM role to authorize model invocation. The backend handles user authentication, request validation, conversation state, and output checks. For retrieval-augmented generation, documents are stored in S3 and ingested into a knowledge base backed by a supported retrieval store, such as an OpenSearch Serverless vector collection. The application retrieves relevant context before generating an answer and applies document-level access restrictions appropriate to its users.

Tool-enabled applications connect agents to narrowly scoped Lambda functions that validate arguments and call business systems. Step Functions is useful when a workflow needs explicit stages, retries, durable execution, or review outside the model interaction. For example, a document pipeline can generate a structured extraction, validate required fields, and send uncertain results to a manual queue. CloudWatch metrics and application telemetry track latency, errors, throttling, and consumption, while optional invocation logs support debugging under controlled access. Keep storage, model execution, retrieval, and action permissions separate, and review regional routing and data handling requirements across the complete request path.

Amazon S3Amazon S3

Use S3 as a document source for supported knowledge base ingestion workflows or as input and output storage for batch inference. For example, upload approved product manuals to a dedicated bucket prefix and synchronize the knowledge base after updates. Scope bucket permissions to the relevant service role and configure encryption and retention for the data.

AWS LambdaAWS Lambda

Lambda functions can invoke Bedrock runtime APIs and implement agent action groups. For example, an order-status action can validate an order identifier, check the caller's authorization, and query an order system. Keep action permissions narrow and make write operations idempotent where retries are possible.

AWS Step FunctionsAWS Step Functions

Use Step Functions to orchestrate Bedrock calls alongside validation, retries, and downstream processing. For example, a document-processing state machine can invoke a model, validate the returned JSON in Lambda, and route invalid results for review. Step Functions provides supported Bedrock service integrations, while other operations can be called through supported SDK integrations or Lambda.

Amazon CloudWatchAmazon CloudWatch

Use CloudWatch metrics to monitor invocation activity, latency, errors, and throttling. For example, alarm on sustained throttling and correlate it with application request volume before requesting quota increases. Optional model invocation logging can capture supported request and response data, so restrict access and set retention policies before enabling it.

Amazon API GatewayAmazon API Gateway

Amazon Bedrock model invocation APIs can be exposed through API Gateway REST APIs using AWS service integrations. API Gateway adds client authentication, request validation, and throttling in front of model inference.

Amazon AuroraAmazon Aurora

Amazon Bedrock Knowledge Bases can store embeddings and document metadata in Amazon Aurora PostgreSQL-Compatible databases using pgvector. Bedrock retrieves relevant passages from Aurora to ground model responses in your documents.

Amazon Aurora ServerlessAmazon Aurora Serverless

Amazon Bedrock Knowledge Bases supports Aurora PostgreSQL-Compatible Serverless v2 as a vector store. This lets the database capacity scale with the ingestion and retrieval workload.

Amazon Bedrock AgentCore

Amazon Bedrock foundation models power AgentCore Memory's extraction and consolidation of long-term memories. AgentCore uses these models to turn agent conversations into reusable information such as user preferences and factual memories.

Amazon ConnectAmazon Connect

Amazon Bedrock powers generative AI capabilities in Amazon Connect, including AI agents that answer questions and assist customer service representatives. Connect can use Bedrock models and knowledge bases to generate responses grounded in an organization's support content.

Amazon Data FirehoseAmazon Data Firehose

Amazon Bedrock Knowledge Bases can deliver ingestion logs to Amazon Data Firehose through CloudWatch log delivery. Firehose can route these records to supported destinations for analysis and operational troubleshooting.

Amazon Bedrock exam questions

Real practice exam questions on Amazon Bedrock, with the correct answer and its explanation. Answer them with every option in the exam simulator.

AIB-C01 (beta)

A wholesale distributor wants to reduce the employee effort required to process order-change requests in its customer portal. Customers submit requests such as, “Replace the unavailable items with compatible products, keep the total below $4,000, and deliver everything by Friday.” Each request can require a different combination of inventory checks, compatibility checks, shipping quotes, and order updates. These systems expose APIs, but no single predefined sequence handles all requests. The distributor has application developers but does not want to train models or operate model-serving infrastructure. Customers must approve proposed changes, and existing systems must enforce pricing and purchasing permissions. Which initiative BEST maps AI capabilities to this business objective?

Show answerCORRECT ANSWER

Use Agents for Amazon Bedrock to interpret requests and coordinate approved APIs for checks and proposed changes. Require customer confirmation and independent system authorization before updates. Measure employee effort per successfully completed request while maintaining order accuracy.

Agent-based tool use maps the customer's variable, natural-language request to the necessary live system operations without requiring the distributor to operate foundation-model serving infrastructure. Customer confirmation and system-enforced authorization preserve control. Measuring employee effort per successful request connects the capability to the business objective.

MLA-C02 (beta)

A company uses a third-party evaluation platform to invoke foundation models in Amazon Bedrock. The platform supports Bedrock bearer-token authentication but cannot sign requests with AWS Signature Version 4. It imports credentials from the company's secure integration once every 24 hours and cannot refresh them between imports. Evaluation requests run continuously.

Security policy requires replacing the platform's credential at each daily import and invalidating its predecessor after successful authentication with the replacement. The integration can provision credentials and update the platform without interrupting requests.

Which credential choice and replacement procedure meet these requirements?

Show answerCORRECT ANSWER

Create a long-term Amazon Bedrock API key with sufficient validity to cover the next import. Securely import it, verify successful model invocation with it, and then revoke the previous API key.

A long-term Bedrock API key supports bearer-token authentication and can remain valid throughout the 24-hour import interval. Keeping the previous key valid until the replacement is verified preserves availability. Revoking the previous key then completes rotation.

SAP-C03

A company is validating an Amazon CloudWatch dashboard before deploying a document-analysis application that uses Amazon Bedrock InvokeModel. The dashboard must distinguish accepted requests from client errors, server errors, and throttles. It must also alarm when server-side failures exceed 5.2% of requests accepted for processing.

A controlled test sends exactly 1,000 requests to one model in one AWS Region, with retries disabled. The requests have these mutually exclusive outcomes:

• 850 complete successfully.
• 50 are accepted for processing and subsequently fail with AWS server-side errors.
• 30 are accepted for processing and subsequently fail with client-side errors.
• 40 are rejected during input validation before processing begins.
• 20 are rejected before processing because of a service quota; Bedrock does not classify these rejections as throttles.
• 10 are throttled before processing begins.

There is no other traffic. All metric data has arrived, and the dashboard uses Sum statistics with matching dimensions over the complete test interval.

Which dashboard configuration correctly reconciles the test results and evaluates the alarm?

Show answerCORRECT ANSWER

Expect Invocations = 930, InvocationClientErrors = 90, InvocationServerErrors = 50, and InvocationThrottles = 10. Calculate the server-side failure percentage as 100 × InvocationServerErrors / Invocations; the alarm breaches its threshold.

Invocations includes the 850 successful requests and all 80 requests that fail after acceptance, giving 930. InvocationClientErrors includes 30 post-acceptance client errors, 40 input-validation rejections, and 20 non-throttle quota rejections, giving 90. The 50 server errors are already included in Invocations. Only the 10 throttled requests count toward InvocationThrottles. The server-side failure percentage is 100 × 50 / 930, approximately 5.38%, which exceeds 5.2%.

Practice Amazon Bedrock in a full exam

Questions on Amazon Bedrock in each practice exam, with every answer option and an explanation for each:

Start a free practice exam

What is Amazon Bedrock

3 min read

Amazon Bedrock provides managed access to foundation models for tasks such as text generation, summarization, embeddings, and image generation. Available capabilities depend on the selected model and AWS Region. Applications call Bedrock runtime APIs using AWS SDKs or HTTPS requests, while AWS operates the underlying inference infrastructure. Model selection remains an application design decision because models differ in supported inputs, context windows, response formats, latency, and pricing.

For conversational applications, the Converse API provides a consistent request structure across supported models, including messages, inference settings, and tool definitions. Model-specific capabilities may require additional parameters or native invocation APIs. Engineers should verify model support before assuming that streaming, tool use, structured responses, or multimodal inputs work identically across providers. Production clients should implement timeouts, bounded retries, throttling handling, and limits on input and output size.

Knowledge Bases for Amazon Bedrock supports retrieval-augmented generation by connecting enterprise data to model requests. A common workflow ingests documents from Amazon S3, divides them into chunks, generates embeddings, and stores searchable representations in a supported vector store. At query time, the application retrieves relevant material and supplies it as grounding context. Retrieval quality depends on document preparation, chunking, metadata filters, embedding selection, and access controls; connecting a knowledge base does not guarantee accurate answers.

Agents for Amazon Bedrock can coordinate model reasoning, knowledge retrieval, and actions against application systems. Action groups describe operations that an agent can invoke, commonly through AWS Lambda or application-controlled execution. Treat generated tool arguments as untrusted input, validate permissions independently, and require confirmation for consequential operations such as issuing refunds or modifying customer records.

Amazon Bedrock AgentCore provides another way to deploy agents using any framework and model, without managing infrastructure. Its capabilities can be adopted together or independently, including connections to enterprise systems with authentication and access controls, and tracing, debugging, and evaluation for agent performance.

Guardrails for Amazon Bedrock provides configurable controls for supported content categories, denied topics, sensitive information, and other checks. These controls complement application validation rather than replacing authorization or business rules. Evaluate guardrails against realistic traffic because filtering behavior and supported checks vary by configuration and workflow.

Secure deployments use IAM permissions, encryption controls, and, where appropriate, private connectivity through supported VPC endpoints. Separate permissions for model invocation, knowledge ingestion, and agent actions. Model availability, quotas, inference profiles, and cross-Region routing require explicit review, particularly when workloads have residency requirements. Monitor application latency, token consumption, throttling, and retrieval effectiveness, and enable model invocation logging only with suitable controls for sensitive prompts and responses. Choose pricing and capacity options per model and workload rather than assuming a single billing model across the service.

Service-tier selection is part of the inference offering, and model pricing depends on modality, provider, and model.

For custom Amazon Nova models developed with parameter-efficient training techniques, inference can use either On Demand or Provisioned Throughput. Full Rank fine-tuned Nova models require Provisioned Throughput; custom-model prices are the same as base-model prices.

For GPT-6 Astra on the bedrock-runtime endpoint, tokens-per-minute quota accounting uses a 10x output-token burndown rate: each output token consumes 10 tokens of quota.

Learn Amazon Bedrock in lessons

35 short lessons on Amazon Bedrock in 7 certification courses.

and 29 more lessons

Open the lessons

Concepts related to Amazon Bedrock

AWS architecture icons are the property of Amazon Web Services. BlowStack is not affiliated with or endorsed by AWS.