Claude inside your AWS account
Amazon Bedrock is AWS's managed service for foundation models, including Claude. Choosing it over the direct Anthropic API is usually about where the work lives: teams already on AWS get Claude without leaving their cloud.
Key concept · What Bedrock changes — and what it doesn't
What changes: access and governance move into AWS — authentication via IAM, usage billed through your AWS account, data staying within your AWS security and compliance boundary, and deployment across AWS regions (with cross-region inference to spread capacity). What doesn't change: it's the same Claude models with the same core capabilities. That's why this course teaches the same arc as the direct-API course — the concepts transfer; only the front door differs.
Calling Claude through the Bedrock client
The request flow is the familiar one, wrapped in AWS: your server uses a Bedrock client to send the user's message plus a model ID to Bedrock; the chosen Claude model generates a response; Bedrock returns it; your server forwards it to the user.
Key concept · The Bedrock client, model IDs, and auth
You call Bedrock with the AWS SDK (e.g. boto3's bedrock-runtime client), authenticated by your AWS credentials/IAM role rather than an Anthropic API key. You name the model by its Bedrock model ID (and must request model access in the Bedrock console first). Two request styles exist: the model-native Messages payload, and Bedrock's unified Converse API that gives one consistent shape across providers. The Anthropic SDK also offers a Bedrock mode so the code you learned on the direct API ports over with minimal change.
Everything else you already know
Because it's the same model, the whole developer toolkit from the API chapters carries over on Bedrock: multi-turn conversations, system prompts, temperature, streaming, structured data, prompt evaluation, prompt engineering, tool use, RAG, and the features of Claude. The course walks all of it again in an AWS context.
Key concept · Transfer your knowledge, mind the AWS specifics
Treat Chapters 19–22 as the substance and this chapter as the adapter. The concepts are identical; the AWS-specific details to keep in mind are the credential model (IAM, not API keys), model IDs and access requests, region/quota management, and AWS-native building blocks you may reach for instead of rolling your own — for example, Bedrock Knowledge Bases as a managed path for RAG. Learn Claude once; deploy it wherever your infrastructure already is.
Chapter 23 quiz
Thirteen questions on Bedrock access, auth, and what transfers. Answer first, then expand Show answer.
Part A · Multiple choice
The main reason to use Bedrock over the direct API is:
- A Teams already on AWS get Claude within their existing cloud, billing, and governance
- B A different, better Claude
- C It's the only way to use tools
- D No reason
Show answer
A. Same Claude, AWS-native access.
How do you authenticate to Claude on Bedrock?
- A AWS credentials / IAM
- B An Anthropic API key
- C A Google service account
- D No auth
Show answer
A. Not an Anthropic API key.
Which client/SDK calls Bedrock from Python?
- A The AWS SDK (boto3's bedrock-runtime), or the Anthropic SDK's Bedrock mode
- B A browser extension
- C The gcloud CLI
- D None
Show answer
A.
You name the model on Bedrock by its:
- A Bedrock model ID (after requesting model access)
- B Email address
- C IP address
- D Temperature
Show answer
A.
Bedrock's unified request shape across providers is the:
- A Converse API
- B Messages-only API
- C Vertex API
- D Files API
Show answer
A — the Converse API.
A managed AWS path for RAG is:
- A Bedrock Knowledge Bases
- B IAM
- C CloudFront
- D The Converse API
Show answer
A.
Part B · True or false
Claude on Bedrock is a different model from Claude on the direct API.
Show answer
False. Same Claude models; only the access surface differs.
Multi-turn, system prompts, tool use, and RAG all transfer to Bedrock.
Show answer
True. The concepts are identical.
You must request model access before calling a Claude model on Bedrock.
Show answer
True. In the Bedrock console.
Usage on Bedrock is billed through your AWS account.
Show answer
True. Within your AWS security/compliance boundary.
Cross-region inference can help spread capacity across AWS regions.
Show answer
True.
Part C · Short answer
Name two things Bedrock changes vs. the direct API.
Show answer
Any two: IAM auth instead of API keys, AWS billing, data in your AWS boundary, region/quota management, model IDs + access requests.
Describe the Bedrock request flow in one line.
Show answer
Server uses the Bedrock client to send the user message + a model ID to Bedrock; the model generates a response; Bedrock returns it; the server forwards it to the user.