AWS Bedrock setup
AWS Bedrock requires IAM policy configuration and an authentication method before you can register it as an LLM provider in Port. Complete the steps below before proceeding to Step 2: Store API Keys in Secrets in the main setup guide.
Step 1: Configure IAM policy
Set up an IAM policy to grant permissions for invoking Bedrock models. Serverless models are automatically available, but you control access through IAM policies. Anthropic models require additional setup — see Anthropic models requirements below.
Option 1: Allow specific models
Restrict access to specific models (recommended). Example for Anthropic models in Europe:
View IAM policy example (click to expand)
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "AllowBedrockInference",
"Effect": "Allow",
"Action": [
"bedrock:InvokeModel",
"bedrock:InvokeModelWithResponseStream"
],
"Resource": [
"arn:aws:bedrock:*:*:inference-profile/eu.anthropic.claude-sonnet-4-20250514-v1:0",
"arn:aws:bedrock:*::foundation-model/anthropic.claude-sonnet-4-20250514-v1:0",
"arn:aws:bedrock:*:*:inference-profile/eu.anthropic.claude-haiku-4-5-20251001-v1:0",
"arn:aws:bedrock:*::foundation-model/anthropic.claude-haiku-4-5-20251001-v1:0",
"arn:aws:bedrock:*:*:inference-profile/eu.anthropic.claude-sonnet-4-5-20250929-v1:0",
"arn:aws:bedrock:*::foundation-model/anthropic.claude-sonnet-4-5-20250929-v1:0",
"arn:aws:bedrock:*:*:inference-profile/eu.anthropic.claude-sonnet-4-6",
"arn:aws:bedrock:*::foundation-model/anthropic.claude-sonnet-4-6",
"arn:aws:bedrock:*:*:inference-profile/eu.anthropic.claude-sonnet-5",
"arn:aws:bedrock:*::foundation-model/anthropic.claude-sonnet-5",
"arn:aws:bedrock:*:*:inference-profile/eu.anthropic.claude-sonnet-5-5",
"arn:aws:bedrock:*::foundation-model/anthropic.claude-sonnet-5-5",
"arn:aws:bedrock:*:*:inference-profile/eu.anthropic.claude-opus-4-5-20251101-v1:0",
"arn:aws:bedrock:*::foundation-model/anthropic.claude-opus-4-5-20251101-v1:0",
"arn:aws:bedrock:*:*:inference-profile/eu.anthropic.claude-opus-4-6-v1",
"arn:aws:bedrock:*::foundation-model/anthropic.claude-opus-4-6-v1"
]
}
]
}
Each model requires two ARN entries: inference-profile and foundation-model. Adjust the region and model as needed.
Option 2: Allow all models
Use a wildcard policy to allow all models. You can still disable specific models using the Create or connect an LLM provider API.
View IAM policy example (click to expand)
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "AllowBedrockInference",
"Effect": "Allow",
"Action": [
"bedrock:InvokeModel",
"bedrock:InvokeModelWithResponseStream"
],
"Resource": [
"arn:aws:bedrock:*:*:inference-profile/*",
"arn:aws:bedrock:*::foundation-model/*"
]
}
]
}
Using guardrails
If you want to use guardrails with your Bedrock models, add the bedrock:ApplyGuardrail action to your IAM policy:
View IAM policy example with guardrails (click to expand)
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "AllowBedrockInference",
"Effect": "Allow",
"Action": [
"bedrock:InvokeModel",
"bedrock:InvokeModelWithResponseStream",
"bedrock:ApplyGuardrail"
],
"Resource": [
"arn:aws:bedrock:*:*:inference-profile/*",
"arn:aws:bedrock:*::foundation-model/*"
]
}
]
}
Anthropic models requirements
One-time usage form
- Submit a one-time usage form through the Amazon Bedrock playground or
PutUserCaseForModelAccessAPI. - For AWS Organizations, complete at the management account level; approval extends to child accounts.
AWS Marketplace subscription
- Some Anthropic models require an AWS Marketplace subscription.
- Subscriptions auto-create on first invocation if IAM includes
aws-marketplace:Subscribe, or an admin can enable models first via console/API.
For details, see the AWS Security Blog post.
Step 2: Choose authentication method
After configuring the IAM policy, choose how Port authenticates with AWS Bedrock:
- Assume role (recommended) — Configure an IAM role that Port's LLM gateway can assume. This eliminates the need to store long-lived credentials.
- Access keys — Store AWS access key ID and secret access key in Port secrets. Configure this in Step 2: Store API Keys in Secrets.
- API key — Store an Amazon Bedrock API key in a Port secret. Port sends it as an
Authorization: Bearerheader.
Use exactly one of these methods in your provider configuration.
Using assume role (recommended)
Trust relationship configuration
Create a trust relationship policy on your IAM role that allows Port's LLM gateway roles to assume it:
View trust relationship policy example (click to expand)
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "Statement1",
"Effect": "Allow",
"Principal": {
"AWS": [
"arn:aws:iam::185657066287:role/port-ai-bring-your-own-llm-eu-west-1",
"arn:aws:iam::185657066287:role/port-ai-bring-your-own-llm-us-east-1"
]
},
"Action": "sts:AssumeRole",
"Condition": {
"StringEquals": {
"sts:ExternalId": "<OPTIONAL_EXTERNAL_ID>"
}
}
}
]
}
The example above includes both the EU (app.port.io) and US (app.us.port.io) gateway roles. This is the recommended approach — it ensures your trust policy works regardless of which Port region handles the request.
The sts:ExternalId condition is optional but recommended for additional security. If you use an external ID, create it as a secret in Port before configuring the provider. See Step 2: Store API Keys in Secrets for instructions.
The secret value stored in Port must match the sts:ExternalId in your trust policy character-for-character. Even a trailing space or newline will cause the assume-role to fail silently. Copy-paste the value directly to avoid mismatches.
Using access keys
Store your AWS access key ID and secret access key in Port secrets. See Step 2: Store API Keys in Secrets for instructions.
Using an API key
Generate an API key in the AWS Console under Amazon Bedrock → API keys. A long-term API key is tied to an IAM user, so the IAM policy from Step 1 still controls which models it can invoke.
Store the key as a Port secret (see Step 2: Store API Keys in Secrets) and reference it with apiKeySecretName:
{
"provider": "bedrock",
"enabled": true,
"config": {
"apiKeySecretName": "aws-bedrock-api-key",
"region": "eu-west-1"
}
}
Bedrock API keys can expire. When you rotate a key, update the secret value in Port. No change to the provider configuration is needed.
Using a custom base URL (proxy or gateway)
By default, Port calls the AWS Bedrock runtime endpoint for your region (https://bedrock-runtime.<region>.amazonaws.com). If you route Bedrock traffic through your own proxy or gateway, set baseUrl to override that endpoint.
Port uses the Bedrock Converse API, so your proxy must forward these routes to Bedrock:
POST {baseUrl}/model/{modelId}/conversePOST {baseUrl}/model/{modelId}/converse-stream
baseUrl supports any of the authentication methods described in Step 2: Choose authentication method.
The region field is still required. Port uses it to build the model ID: regions starting with eu use the eu. inference profile prefix (for example, eu.anthropic.claude-sonnet-5), and all other regions use us.. Make sure your proxy accepts these model IDs.
View proxy configuration example (click to expand)
This example routes requests through a LiteLLM Bedrock pass-through endpoint, authenticating with a LiteLLM virtual key stored in the LITELLM_API_KEY secret:
{
"provider": "bedrock",
"enabled": true,
"config": {
"baseUrl": "https://litellm.example.com/bedrock",
"apiKeySecretName": "LITELLM_API_KEY",
"region": "eu-west-1"
}
}