> ## Documentation Index
> Fetch the complete documentation index at: https://docs.datafold.com/llms.txt
> Use this file to discover all available pages before exploring further.

# AWS Bedrock

Connect to Anthropic (Claude) models served from your AWS account via Amazon Bedrock.

**Steps to complete:**

1. [Create an IAM identity with Bedrock access](/integrations/llm-providers/aws-bedrock#create-an-iam-identity)
2. [Configure in Datafold](/integrations/llm-providers/aws-bedrock#configure-in-datafold)

## Create an IAM identity

Create (or reuse) an IAM user with programmatic access and attach a policy allowing:

* `bedrock:InvokeModel` and `bedrock:InvokeModelWithResponseStream` — inference
* `bedrock:ListFoundationModels` — used by Datafold's automatic model listing

Generate an **access key ID / secret access key** pair for it.

## Configure in Datafold

1. Navigate to **Settings** → **Integrations**, click **Add New Integration**, and choose **AWS Bedrock**.
2. Fill in:

| Field                                 | Description                                                                                   |
| ------------------------------------- | --------------------------------------------------------------------------------------------- |
| **Region**                            | The AWS region hosting Bedrock, e.g. `us-east-1`.                                             |
| **Access key ID / Secret access key** | The IAM key pair. Requests are SigV4-signed against the native Bedrock API.                   |
| **Models**                            | Anthropic models available in the region are listed automatically. Select the ones to expose. |

3. Click **Save**.

<Info>
  Newer Claude models that require cross-region inference profiles are handled automatically — Datafold prepends the regional profile prefix (`us.`/`eu.`) at request time.
</Info>

<Info>
  A **Bedrock API key** (bearer token) is also accepted in place of the key pair, but it dispatches over Bedrock's OpenAI-compatible endpoint, which serves only the OpenAI open-weight model family — the access-key pair is the recommended path for Claude.
</Info>