How to route Claude through AWS Bedrock with kRouter

If you work at a company with AWS as the primary cloud, your compliance team probably wants you to route Claude through AWS Bedrock rather than calling Anthropic directly. Bedrock gives you:

• Region-pinned model invocations (data residency)
• PrivateLink so traffic never crosses the public internet
• Inclusion in your AWS bill (one invoice, one procurement)
• Native IAM-based access control

kRouter speaks Bedrock natively. Here is the setup.

Step 1: Enable Claude in your AWS region

Bedrock requires model access requests per region. In AWS Console:

1. Bedrock → Model Access → Manage model access
2. Request access to Anthropic models in your region (us-east-1, eu-west-1, etc.)
3. Approval is usually instant for Sonnet, may take a day for Opus

Step 2: Get IAM credentials

Create an IAM user (or role) with bedrock:InvokeModel and bedrock:InvokeModelWithResponseStream permissions on the Claude model ARNs.

{
  "Version": "2012-10-17",
  "Statement": [{
    "Effect": "Allow",
    "Action": [
      "bedrock:InvokeModel",
      "bedrock:InvokeModelWithResponseStream"
    ],
    "Resource": "arn:aws:bedrock:us-east-1::foundation-model/anthropic.*"
  }]
}

Generate an access key pair for the IAM user.

Step 3: Add Bedrock as a provider in kRouter

Open http://localhost:20128/dashboard:

1. Providers → Add Provider → AWS Bedrock
2. Region: us-east-1 (or your region)
3. Access Key ID: AKIA...
4. Secret Access Key: wJalr...
5. Save & Test

kRouter validates the connection and lists the Claude models you have access to:

bedrock/anthropic.claude-opus-4-7
bedrock/anthropic.claude-sonnet-4-6
bedrock/anthropic.claude-haiku-4-5

Step 4: Use Bedrock in your IDE

Configure your IDE to point at kRouter:

OPENAI_BASE_URL=http://localhost:20128/v1
OPENAI_API_KEY=sk-krouter-local

Then in your model selector, pick bedrock/anthropic.claude-sonnet-4-6. Your IDE sends OpenAI-format requests; kRouter translates them to Bedrock's protocol and authenticates with AWS SigV4 on the wire.

How the format translation works

This is the part most guides skip. Bedrock's Claude API is not the same as Anthropic's direct Messages API, and neither is the same as the OpenAI Chat Completions format your IDE speaks. kRouter handles a three-way translation:

Your IDE sends (OpenAI format):

{
  "model": "bedrock/anthropic.claude-sonnet-4-6",
  "messages": [{"role": "user", "content": "refactor this function"}],
  "stream": true
}

kRouter translates to (Bedrock Anthropic format):

{
  "anthropic_version": "bedrock-2023-05-31",
  "messages": [{"role": "user", "content": "refactor this function"}],
  "max_tokens": 4096
}

The key differences kRouter handles:

• Model ID mapping. Your IDE says bedrock/anthropic.claude-sonnet-4-6; kRouter maps that to the Bedrock model ARN anthropic.claude-sonnet-4-6-v2:0.
• Auth injection. The outbound request to Bedrock uses AWS SigV4 signing with your IAM credentials. Your IDE never sees the AWS credentials -- it just uses sk-krouter-local.
• Streaming translation. Bedrock uses a custom binary event-stream format (application/vnd.amazon.eventstream), not the standard SSE your IDE expects. kRouter decodes the binary frames, extracts the JSON deltas, and re-encodes them as OpenAI-compatible data: {...} SSE lines.
• Stop reason mapping. Bedrock returns stop_reason: "end_turn" in its own format; kRouter normalizes it to OpenAI's finish_reason: "stop".
• Token counting. Bedrock includes usage metadata in a different response field. kRouter extracts it and emits standard usage: { prompt_tokens, completion_tokens } so your dashboard metrics stay consistent across providers.

This is one of the 9 format translations kRouter supports. The same translation engine handles OpenAI, Claude, Gemini, Cursor, Kiro, Vertex, Antigravity, Ollama, and the Responses API. See the /compare page for a full matrix.

Step 5 (optional): Build a compliance-friendly combo

1. bedrock/anthropic.claude-sonnet-4-6   # Primary, enterprise-compliant
2. anthropic/claude-sonnet-4-6           # Fallback, only if Bedrock fails
3. kr/claude-sonnet-4.5                  # Free fallback for non-sensitive tasks

This setup keeps sensitive workloads on Bedrock (compliant region, PrivateLink), with a controlled fallback chain.

Why route Bedrock through kRouter at all?

You could call Bedrock directly. Why add kRouter?

• One endpoint for all providers. Your IDE sees localhost:20128/v1 whether you are using Bedrock, Anthropic direct, OpenAI, or Kiro. Switching is a model-name change, not a config overhaul.
• Format translation. Bedrock's Claude API is not the same as Anthropic's direct Claude API. kRouter handles the translation transparently -- including the binary streaming format.
• RTK compression. Bedrock charges by input token. RTK saves 20-40% on those bills because it compresses tool outputs before they reach the billing meter.
• Multi-region failover. Configure Bedrock in two regions. When one has capacity issues, kRouter falls over to the other. The Zenith Score Engine ranks regions by live latency and quota headroom, so the healthiest region always wins.

A real Bedrock combo for a regulated enterprise

1. bedrock-us-east-1/anthropic.claude-sonnet-4-6   # Primary US
2. bedrock-eu-west-1/anthropic.claude-sonnet-4-6   # EU failover
3. anthropic/claude-sonnet-4-6                     # Last resort (direct API)

Compliant. Resilient. Cost-controlled. One IDE config.

Install

npm install -g @sifxprime/krouter
krouter -t

Bedrock setup guide at /install. Full provider catalog at /providers. Changelog at /changelog.