Skip to main content
Amazon Bedrock lets you run model inference entirely within your AWS account. Choose this provider if your team is already on AWS, you have Bedrock enabled, or you have compliance requirements that mandate keeping LLM traffic within your cloud boundary.
Bedrock is the LLM provider option that requires AWS — the MergeWatch server itself can still run anywhere (Docker, GCP, Azure, bare metal). Only the LLM API calls go to Bedrock.

​
Configuration

Set these environment variables in your .env file or container environment:
If MergeWatch runs on EC2, ECS, or Fargate, you can use an IAM instance profile or task role instead of static credentials. The AWS SDK picks up credentials automatically — just omit AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY.

​
Default model

MergeWatch uses us.anthropic.claude-sonnet-4-6 for review agents by default, and us.anthropic.claude-haiku-4-5-20251001-v1:0 as the lightModel for low-complexity tasks such as summary generation. Both are cross-region inference profiles that route requests to the nearest available region. Override either with model: / lightModel: in .mergewatch.yml — see the configuration reference.

​
Bedrock model access

Before using a model, you must enable it in the Bedrock console:
  1. Open the Bedrock Model Access console
  2. Find the Claude model you want to use
  3. Click Request model access if the status is not Access granted
  4. Wait for approval (usually instant for Anthropic models)
If you skip this step, MergeWatch will receive AccessDeniedException errors from Bedrock. Check the Troubleshooting guide if you see these errors.

​
Supported regions

​
Cross-region inference profiles

Bedrock supports cross-region inference profiles that automatically route requests to the nearest region with available capacity. These profiles use a geographic prefix: The default model us.anthropic.claude-sonnet-4-6 uses the us. prefix, routing across US regions. If your deployment is in Europe or Asia Pacific, swap the prefix for the matching geography.
Not every model is available as an inference profile in every geography, and the availability list changes over time. Before setting a prefixed ID, confirm the exact profile exists in your account — the Bedrock Model Access console lists the inference profile IDs you can actually call. An ID that is valid in us. may not exist under eu. or ap..

​
Model override examples

​
Next steps

Anthropic (direct)

The default provider — simplest setup, no AWS required.

Configure review behavior

Tune sensitivity, ignored paths, and review focus areas.

Environment variables

Full list of supported environment variables.

Platform guides

Deploy MergeWatch on AWS ECS Fargate.