Skip to content
Nubent

/Glossary · AI

What is Amazon Bedrock?

Amazon Bedrock is a fully managed AWS service that gives you access to foundation models from providers like Anthropic, Meta, Mistral AI and Amazon through a single API. It includes the tooling to build generative AI applications and agents with enterprise-grade security and privacy.

How Amazon Bedrock works

Bedrock exposes text, image and embedding models through APIs like Converse and InvokeModel, so you never manage servers or GPUs. On-demand, you pay per token processed, and there are batch inference and provisioned throughput options for workloads with predictable volume.

On top of the models, Bedrock adds production-ready building blocks: Knowledge Bases for RAG, Guardrails to filter harmful content and sensitive data, model evaluation, prompt management, and Bedrock AgentCore to deploy and run agents. Your data is not used to train the base models, and you can reach the service through private VPC endpoints.

When Amazon Bedrock makes sense

It fits when your application already runs on AWS and you need generative AI with enterprise controls: IAM, CloudTrail, KMS encryption and private connectivity. It also fits when you want to switch models without rewriting the integration, comparing quality, latency and cost across providers behind the same API.

If you need to train your own model from scratch or control every detail of inference, Amazon SageMaker AI may be a better fit. For most business use cases, such as assistants, document search, data extraction or agents, Bedrock cuts a lot of the work between prototype and production.

Best practices and common mistakes

Build your own evaluation set before picking a model, turn on model invocation logging to CloudWatch, and define Guardrails before anything reaches users. Use cross-Region inference profiles to improve availability, and consider prompt caching and smaller models for simple tasks.

Common mistakes: using the largest model for everything and watching costs spike, going to production without quality metrics, and sending far too much context on every call. Another one is treating the prompt as static, when it needs versioning and testing like any other code.

/FAQ

Questions about Amazon Bedrock

Another question? Write to hello@nubent.com and an engineer will answer.

Which models are available in Amazon Bedrock?

Bedrock offers models from Anthropic (Claude), Meta (Llama), Mistral AI, Cohere, Amazon Nova and other providers. The catalog changes often and availability varies by Region, so check the console or the documentation before you design the solution.

Does Amazon Bedrock use my data to train models?

No. AWS states that prompts and responses are not used to train the base models and are not shared with model providers. Data is encrypted in transit and at rest, and you can use AWS PrivateLink so traffic never goes over the public internet.

How much does Amazon Bedrock cost?

On-demand, you pay for input and output tokens, with prices that vary by model. Batch inference is discounted, and provisioned throughput is available. Features like Knowledge Bases and Guardrails are billed separately, plus whatever vector store you choose.

What is the difference between Bedrock and SageMaker?

Bedrock gives you foundation models ready to call through an API, with no infrastructure to manage. Amazon SageMaker AI is a platform to train, fine-tune and deploy your own machine learning models with full control. Many companies use Bedrock for generative AI and SageMaker for predictive models.

Next step

What should we build next on AWS?

30 minutes with an engineer. Concrete next steps, whether or not you work with us.