Skip to main content

​
Installation

​
Overview

The @ai-billing/amazon-bedrock package provides middleware for tracking token usage and calculating costs when using Amazon Bedrock models with the Vercel AI SDK. Amazon Bedrock’s Converse API reports usage with inputTokens/outputTokens field names (camelCase, no underscore) instead of the prompt_tokens/completion_tokens naming used by OpenAI-style providers. Bedrock’s inputTokens also excludes cache-read and cache-write tokens — they are reported separately as cacheReadInputTokens/cacheWriteInputTokens — so billing adds cache costs on top of the full prompt cost instead of deducting them, the same accounting style used by the Anthropic middleware.

​
Usage

To use the middleware, wrap your Amazon Bedrock model using wrapLanguageModel from the ai package and pass the createAmazonBedrockMiddleware.
1

Initialize the Amazon Bedrock provider

First, set up the provider using @ai-sdk/amazon-bedrock with your AWS credentials and region.
2

Define model pricing

Set up a price resolver to define the costs for the models you’ll be using. Note that a cross-region inference profile id (e.g. us.amazon.nova-lite-v1:0) is billed at its own rate, distinct from the bare model id, so key your pricing map by the exact id you invoke.
3

Create the billing middleware

Initialize the Amazon Bedrock billing middleware. You need to provide a destination (such as consoleDestination) where billing events will be sent, along with your priceResolver.
4

Wrap the model

Use wrapLanguageModel from the ai package to apply the billing middleware to your Amazon Bedrock model.
5

Use the wrapped model

Finally, use the wrapped model with AI SDK functions like generateText or streamText. The billing middleware will automatically track tokens and calculate costs.