Claude Platform on AWS

The Claude Platform on AWS provider for the AI SDK gives you access to the Anthropic Messages API hosted inside your AWS environment. Anthropic operates the stack, so you get the same API surface and feature set as the first-party Anthropic provider — including streaming, prompt caching, tool use, computer use, Agent Skills, and anthropic-beta headers — with AWS-native authentication and AWS Marketplace billing. Start with AI Gateway for models in its catalog (recommended), or use @ai-sdk/anthropic-aws to connect directly to Claude Platform on AWS.

import { generateText } from 'ai';

const { text } = await generateText({
  model: 'anthropic/claude-sonnet-5.5',
  prompt: 'Write a vegetarian lasagna recipe for 4 people.',
});

console.log(text);

Using AI Gateway

The AI SDK uses AI Gateway by default for model strings. Gateway provides access to models across providers through a single API key, with usage and cost tracking. Install ai:

pnpm add ai

npm install ai

yarn add ai

bun add ai

Set an AI Gateway API key in your environment:

export AI_GATEWAY_API_KEY=your_api_key_here

Vercel deployments can also use OIDC authentication. See the AI Gateway provider documentation for configuration and the model catalog for available models.

The example sets providerOptions.gateway.order to ['claudeaws']. This sends the request to Claude Platform on AWS first, then falls back to other available providers of the same model if Claude Platform on AWS is unavailable or the request fails. order sets a preference; it does not restrict fallback to Claude Platform on AWS. See provider ordering.

The dropdown lists models served by Claude Platform on AWS through Gateway with a corresponding direct-provider example. Provider availability depends on the model and your team's Provider Allowlist.

Gateway uses the claudeaws route for Claude Platform on AWS. Enable claudeaws in your team's Gateway Provider Allowlist if it is disabled; otherwise order allows Gateway to fall back to another provider. It authenticates the upstream request using Gateway-managed credentials; your AWS workspace and AWS Marketplace billing are configured separately when using the direct provider.

Using @ai-sdk/anthropic-aws

For direct access to Claude Platform on AWS, install ai and @ai-sdk/anthropic-aws:

pnpm add ai @ai-sdk/anthropic-aws

npm install ai @ai-sdk/anthropic-aws

yarn add ai @ai-sdk/anthropic-aws

bun add ai @ai-sdk/anthropic-aws

Set your Claude Platform on AWS API credential in the ANTHROPIC_AWS_API_KEY environment variable:

export ANTHROPIC_AWS_API_KEY=your_api_key_here
export AWS_REGION=us-west-2
export ANTHROPIC_AWS_WORKSPACE_ID=your_workspace_id

Use the direct provider when you need AWS-native authentication, AWS Marketplace billing, or access to your Claude Platform workspace. Gateway requests do not use your workspace configuration.

This differs from Amazon Bedrock in two important ways: Claude Platform on AWS uses Anthropic's Messages API directly (not Bedrock's Converse/InvokeModel), and new features are available the same day they launch on the first-party Claude API (no AWS integration delay).

Prerequisites

Before you can use the provider, your AWS account must be subscribed to Claude Platform on AWS through the AWS Marketplace, and outbound web identity federation must be enabled on the account (a one-time setup step):

aws iam enable-outbound-web-identity-federation

Without this, every request returns Outbound web identity federation is disabled for your account. This is the most common setup error.

You'll also need your workspace ID. When you subscribe, AWS provisions an initial workspace for your account in the selected region. You can find the ID in the Claude Console under Workspaces (accessed from the AWS Console via the Claude Platform on AWS service page).

Authentication

The provider supports two authentication methods:

Using AWS SigV4 (recommended for production)

SigV4 integrates with your existing AWS IAM policies, roles, and auditing. Configure AWS credentials using any method supported by the AWS default credential provider chain — environment variables, shared credentials file, web identity (IRSA), ECS container credentials, or EC2 instance metadata.

AWS_REGION=us-west-2
ANTHROPIC_AWS_WORKSPACE_ID=wrkspc_…
AWS_ACCESS_KEY_ID=…
AWS_SECRET_ACCESS_KEY=…
# AWS_SESSION_TOKEN=…  # only for temporary credentials (SSO, STS, assumed role)

Then use the default provider instance:

import { anthropicAws } from '@ai-sdk/anthropic-aws';

Or instantiate explicitly:

import { createAnthropicAws } from '@ai-sdk/anthropic-aws';

const anthropicAws = createAnthropicAws({
  region: 'us-west-2',
  workspaceId: 'wrkspc_…',
});

For dynamic credentials (e.g., assuming a role at request time):

const anthropicAws = createAnthropicAws({
  region: 'us-west-2',
  workspaceId: 'wrkspc_…',
  credentialProvider: async () => fetchCredentialsFromSTS(),
});

Using an API key

For simpler integration paths (local development, scripts, migration from the first-party Claude API), you can authenticate with an API key instead of SigV4. Your Anthropic account representative provisions API keys for Claude Platform on AWS.

ANTHROPIC_AWS_API_KEY=sk-…

When apiKey is set, it takes precedence over any SigV4 credentials in the environment.

import { createAnthropicAws } from '@ai-sdk/anthropic-aws';

const anthropicAws = createAnthropicAws({
  region: 'us-west-2',
  workspaceId: 'wrkspc_…',
  apiKey: 'sk-…',
});

Provider Settings

Language Models

You can create models that call the Anthropic Messages API on AWS using the provider instance:

import { anthropicAws } from '@ai-sdk/anthropic-aws';

const model = anthropicAws('claude-sonnet-5-5');

Model IDs are identical to the first-party Anthropic API. See the Anthropic provider docs for the full list of language model options, prompt caching, computer use, web search, code execution, and Agent Skills. All of these work identically here because Claude Platform on AWS uses Anthropic's runtime directly.

Example

import { anthropicAws } from '@ai-sdk/anthropic-aws';
import { generateText } from 'ai';

const { text } = await generateText({
  model: anthropicAws('claude-sonnet-5-5'),
  prompt: 'Invent a new holiday and describe its traditions.',
});

IAM permissions

Your IAM principal needs permission to call the Claude Platform on AWS actions on your workspace. AWS provides three managed policies:

• AnthropicFullAccess — grants aws-external-anthropic:* on all resources.
• AnthropicInferenceAccess — grants read actions plus CreateInference, CreateBatchInference, CancelBatchInference, DeleteBatchInference, and CountTokens on all workspaces. This is the minimum for calling models.
• AnthropicReadOnlyAccess — grants Get*, List*, and CallWithBearerToken on all workspaces. Insufficient for inference.