What is Amazon Bedrock?
Amazon Bedrock is AWS's managed AI service. It provides access to a variety of AI models from different vendors through a single AWS interface, with no separate accounts or API keys required. Because it runs inside Kearney's AWS environment, all data stays within the same encrypted, access-controlled infrastructure as the rest of the sandbox.
Bedrock is available for internal Kearney experimentation and prototypes that are not scoped to DoD clients. For any work involving a DoD engagement or federal end users, you must use AskSage instead. This is a firm policy requirement under FASCSA and FAR 52.204-30, not a preference.
Available models
The following model families are enabled in the sandbox. Anthropic (Claude) models are not available. They were removed in February 2026 per a DoD supply chain risk designation.
| Model Family | Provider | Best For |
|---|---|---|
| Amazon Nova | Amazon | General analysis, text generation, AWS-integrated workflows |
| Amazon Titan | Amazon | Embeddings (turning text into searchable vectors), basic text tasks |
| Meta Llama | Meta | Open-source reasoning, coding assistance, research experimentation |
| Mistral | Mistral AI | Fast, efficient responses for lighter analysis tasks |
| Cohere | Cohere | Document search, retrieval, and classification tasks |
How to use Bedrock from your notebook
Bedrock is called from your SageMaker Studio notebook using the AWS Python SDK (boto3). Your notebook session automatically has the necessary permissions. No additional credentials are required.
The sandbox uses the Bedrock Converse API, which provides a consistent interface across all model families. This means you can switch between Nova, Llama, or Mistral by changing only the model name.
Ask your team lead for the Bedrock starter notebook, which includes working examples for common tasks including text generation, summarization, and embedding creation.
Use Bedrock when you are experimenting with AWS-native features, building something that will eventually integrate with other AWS services, or working on an internal Kearney prototype. Use AskSage for anything that will touch a federal client deliverable or a DoD engagement.
What Bedrock cannot do in this environment
Bedrock access in the sandbox is scoped to the approved model families listed above. The IAM policy intentionally excludes Anthropic models. Attempting to call an excluded model will return an access denied error. This is expected behavior, not a bug.
Bedrock does not have a browser interface in this environment. It is only accessible from your SageMaker Studio notebook via the Python SDK. There is no console UI for running prompts directly.
Contact sandbox-support@kearneyco.com. Include the model name you are attempting to call and the error message you received.