---
title: "AWS"
description: "Control AWS credential retrieval for AI agents"
url: "https://gazebohq.com/services/aws"
---

The AWS preset stores an access key ID, secret access key and optional region. Grant retrieval of these values only to agents that need the AWS account or environment. This is credential storage and retrieval, not AWS role assumption or an API proxy.

## Separate agent retrieval grants from AWS permissions

AWS IAM policies determine what a delivered credential can do. Use a dedicated identity with narrowly scoped policies rather than root credentials, and keep production and development credentials separate. Gazebo records retrieval attempts against the named agent; AWS CloudTrail and service-specific logs remain the evidence for subsequent AWS activity. Stopping retrieval does not disable an AWS key already held by a runtime.

## How it works

1. Create a dedicated AWS identity and restrict its IAM policies to the resources and operations required by the task.
2. Choose AWS in Gazebo and store its access key ID and secret access key, with the region if needed.
3. Grant the intended agent retrieval of the credential fields it needs. The runtime uses an AWS client to authenticate directly to AWS.
4. Review Gazebo retrieval records alongside CloudTrail where available. Disable or rotate the AWS access key in AWS if a delivered copy may be exposed.

## Use cases

**Read-only inventory in one account:** Give an inventory agent a dedicated AWS credential with provider-side read permissions for the relevant account and resources. Gazebo identifies which agent retrieved the credential, while AWS determines what it can read.

**Separate production and staging access:** Store credentials for each environment separately and grant agents only the intended environment's fields. Retiring an agent stops new retrieval; disabling its retained AWS key requires action in AWS.
