AWS Bedrock Free code examples in C#, Python, Java and JavaScript

No Bedrock-specific package and no API key. boto3 picks up whatever the AWS credential chain finds — environment variables, ~/.aws/credentials, an EC2 or ECS role. Enable the models you want in the Bedrock console first; a fresh account has none.

pip install boto3

# Any one of these works. A role is preferred in production.
export AWS_REGION="us-east-1"
export AWS_ACCESS_KEY_ID="..."
export AWS_SECRET_ACCESS_KEY="..."

Bedrock ids are per model AND per region, and newer models require an inference profile with a regional prefix (us., eu., apac.). Guessing produces ValidationException or AccessDeniedException, neither of which says 'wrong region'.

import boto3

# Note: bedrock, not bedrock-runtime. Listing is the control plane.
control = boto3.client("bedrock", region_name="us-east-1")

# Most current models must be called through an inference profile.
for profile in control.list_inference_profiles()["inferenceProfileSummaries"]:
    print(profile["inferenceProfileId"], "-", profile["inferenceProfileName"])

# Older models can be called by their plain id.
for model in control.list_foundation_models()["modelSummaries"]:
    if "TEXT" in model.get("outputModalities", []):
        print(model["modelId"])

# Put the one you want in configuration — it differs per environment.

content is a LIST of blocks, not a string; that is what lets the same call carry text, images and documents. Swap modelId for a Llama or Nova profile and nothing else changes.

import os
import boto3

client = boto3.client("bedrock-runtime", region_name=os.environ["AWS_REGION"])

# Example id from the AWS docs — list yours with the step above.
MODEL_ID = os.environ.get(
    "BEDROCK_MODEL_ID", "us.anthropic.claude-3-haiku-20240307-v1:0")

response = client.converse(
    modelId=MODEL_ID,
    system=[{"text": "You are a concise technical assistant."}],
    messages=[{
        "role": "user",
        "content": [{"text": "Explain a database index in two sentences."}],
    }],
    inferenceConfig={"maxTokens": 500, "temperature": 0.2},
)

print(response["output"]["message"]["content"][0]["text"])
print(response["usage"]["inputTokens"], response["usage"]["outputTokens"])
print(response["stopReason"])

converse_stream returns an event dict per chunk. The three failures below all report something other than their cause.

stream = client.converse_stream(
    modelId=MODEL_ID,
    messages=[{"role": "user", "content": [{"text": "Write a haiku about indexes."}]}],
)

for event in stream["stream"]:
    if "contentBlockDelta" in event:
        print(event["contentBlockDelta"]["delta"]["text"], end="", flush=True)

print()

# 1. AccessDeniedException on a real model id
#    Either the model is not enabled for your account in that region
#    (Bedrock console -> Model access), or the IAM policy is missing
#    bedrock:InvokeModel. The message does not distinguish them.

# 2. ValidationException: "model identifier is invalid"
#    Usually a plain model id where an inference profile is required,
#    or the right id in the wrong region.

# 3. ThrottlingException immediately, with no traffic
#    Bedrock quotas are per model, per region, and start low. Raise
#    them in Service Quotas; retrying will not help.

AWS SDK v2. No API key — the default credentials provider chain reads the environment, the shared profile, or an attached role.

<dependency>
    <groupId>software.amazon.awssdk</groupId>
    <artifactId>bedrockruntime</artifactId>
    <version>2.32.0</version>
</dependency>
<dependency>
    <groupId>software.amazon.awssdk</groupId>
    <artifactId>bedrock</artifactId>
    <version>2.32.0</version>
</dependency>

Ids are per model and per region. Two clients: BedrockClient lists, BedrockRuntimeClient invokes.

import software.amazon.awssdk.services.bedrock.BedrockClient;
import software.amazon.awssdk.services.bedrock.model.ListInferenceProfilesRequest;

try (BedrockClient control = BedrockClient.create()) {
    control.listInferenceProfiles(ListInferenceProfilesRequest.builder().build())
        .inferenceProfileSummaries()
        .forEach(profile -> System.out.println(
            profile.inferenceProfileId() + " - " + profile.inferenceProfileName()));
}

ContentBlock.fromText builds the block. Swap the modelId for a Llama or Nova profile and nothing else changes.

import software.amazon.awssdk.services.bedrockruntime.BedrockRuntimeClient;
import software.amazon.awssdk.services.bedrockruntime.model.ContentBlock;
import software.amazon.awssdk.services.bedrockruntime.model.ConversationRole;
import software.amazon.awssdk.services.bedrockruntime.model.ConverseResponse;
import software.amazon.awssdk.services.bedrockruntime.model.Message;
import software.amazon.awssdk.services.bedrockruntime.model.SystemContentBlock;

BedrockRuntimeClient client = BedrockRuntimeClient.create();

// Example id from the AWS docs — list yours with the step above.
String modelId = System.getenv().getOrDefault(
    "BEDROCK_MODEL_ID", "us.anthropic.claude-3-haiku-20240307-v1:0");

Message message = Message.builder()
    .role(ConversationRole.USER)
    .content(ContentBlock.fromText("Explain a database index in two sentences."))
    .build();

ConverseResponse response = client.converse(request -> request
    .modelId(modelId)
    .system(SystemContentBlock.fromText("You are a concise technical assistant."))
    .messages(message)
    .inferenceConfig(config -> config.maxTokens(500).temperature(0.2F)));

System.out.println(response.output().message().content().get(0).text());
System.out.println(response.usage().inputTokens() + " in, "
    + response.usage().outputTokens() + " out");

Streaming uses the ASYNC client — BedrockRuntimeAsyncClient — with a response handler. The three failures below each report something other than their cause.

import software.amazon.awssdk.services.bedrockruntime.BedrockRuntimeAsyncClient;
import software.amazon.awssdk.services.bedrockruntime.model.ContentBlock;
import software.amazon.awssdk.services.bedrockruntime.model.ConversationRole;
import software.amazon.awssdk.services.bedrockruntime.model.ConverseStreamResponseHandler;
import software.amazon.awssdk.services.bedrockruntime.model.Message;

BedrockRuntimeAsyncClient async = BedrockRuntimeAsyncClient.create();

Message prompt = Message.builder()
    .role(ConversationRole.USER)
    .content(ContentBlock.fromText("Write a haiku about indexes."))
    .build();

ConverseStreamResponseHandler handler = ConverseStreamResponseHandler.builder()
    .subscriber(ConverseStreamResponseHandler.Visitor.builder()
        .onContentBlockDelta(chunk -> System.out.print(chunk.delta().text()))
        .build())
    .build();

async.converseStream(r -> r.modelId(modelId).messages(prompt), handler).join();

// 1. AccessDeniedException — model not enabled in that region, OR the
//    IAM policy lacks bedrock:InvokeModel. Same message for both.
// 2. ValidationException "model identifier is invalid" — a plain id
//    where an inference profile is required, or the wrong region.
// 3. ThrottlingException with no traffic — quotas are per model, per
//    region, and start low. Raise them in Service Quotas.

No API key. The SDK uses the AWS credential chain — environment, shared config, or an attached role. Enable model access in the Bedrock console first.

npm install @aws-sdk/client-bedrock-runtime @aws-sdk/client-bedrock

export AWS_REGION="us-east-1"
export AWS_ACCESS_KEY_ID="..."
export AWS_SECRET_ACCESS_KEY="..."

Ids are per model and per region, and current models need an inference profile with a regional prefix. Two different clients: bedrock lists, bedrock-runtime invokes.

import { BedrockClient, ListInferenceProfilesCommand } from "@aws-sdk/client-bedrock";

const control = new BedrockClient({ region: process.env.AWS_REGION });

const { inferenceProfileSummaries } = await control.send(
  new ListInferenceProfilesCommand({}),
);

for (const profile of inferenceProfileSummaries) {
  console.log(profile.inferenceProfileId, "-", profile.inferenceProfileName);
}

// Put the one you want in configuration — it differs per environment.

content is an array of blocks, not a string. Swap modelId for a Llama or Nova profile and nothing else in this call changes.

import {
  BedrockRuntimeClient,
  ConverseCommand,
} from "@aws-sdk/client-bedrock-runtime";

const client = new BedrockRuntimeClient({ region: process.env.AWS_REGION });

// Example id from the AWS docs — list yours with the step above.
const MODEL_ID = process.env.BEDROCK_MODEL_ID
  ?? "us.anthropic.claude-3-haiku-20240307-v1:0";

const response = await client.send(new ConverseCommand({
  modelId: MODEL_ID,
  system: [{ text: "You are a concise technical assistant." }],
  messages: [{
    role: "user",
    content: [{ text: "Explain a database index in two sentences." }],
  }],
  inferenceConfig: { maxTokens: 500, temperature: 0.2 },
}));

console.log(response.output.message.content[0].text);
console.log(response.usage.inputTokens, response.usage.outputTokens);

ConverseStreamCommand returns an async iterable of events. The three failures below each report something other than their cause.

import { ConverseStreamCommand } from "@aws-sdk/client-bedrock-runtime";

const response = await client.send(new ConverseStreamCommand({
  modelId: MODEL_ID,
  messages: [{ role: "user", content: [{ text: "Write a haiku about indexes." }] }],
}));

for await (const event of response.stream) {
  if (event.contentBlockDelta) {
    process.stdout.write(event.contentBlockDelta.delta.text);
  }
}

// 1. AccessDeniedException — the model is not enabled for your account
//    in that region, OR the IAM policy lacks bedrock:InvokeModel. The
//    message does not tell you which.

// 2. ValidationException "model identifier is invalid" — usually a
//    plain model id where an inference profile is required, or the
//    right id in the wrong region.

// 3. ThrottlingException with no traffic — Bedrock quotas are per
//    model, per region, and start low. Raise them in Service Quotas.