> ## Documentation Index
> Fetch the complete documentation index at: https://gomodel-docs-providers-restructure.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Amazon Bedrock

> Configure Amazon Bedrock in GoModel using AWS credentials and the Bedrock Runtime Converse API.

GoModel routes Bedrock requests through the Bedrock Runtime Converse and
ConverseStream APIs, which present a uniform request and response shape across
the model families Bedrock hosts (Anthropic, Amazon Nova, Meta Llama, Mistral,
Cohere, AI21, and others). Responses are normalized into OpenAI-compatible
chat completions.

Flow:

`Client -> GoModel -> Amazon Bedrock`

## Before you start

* An AWS account with Bedrock access in your chosen region.
* Model access granted for the foundation models you want to call (Bedrock
  console → **Model access**).
* AWS credentials reachable through the standard AWS credential chain: env
  vars, `AWS_PROFILE`, IAM Identity Center, container or instance roles.

Bedrock has no API key of its own. GoModel does not store AWS credentials —
they are resolved at runtime by the AWS SDK.

## 1. Set the AWS region

Set `BEDROCK_BASE_URL` to either an AWS region (recommended) or a fully
qualified Bedrock endpoint URL:

```bash theme={null}
export BEDROCK_BASE_URL="us-east-1"
```

When `BEDROCK_BASE_URL` is empty, GoModel falls back to `AWS_REGION` or
`AWS_DEFAULT_REGION` from the standard AWS environment.

## 2. Make AWS credentials available

Any method the AWS SDK supports works. The simplest for local testing:

```bash theme={null}
export AWS_ACCESS_KEY_ID="..."
export AWS_SECRET_ACCESS_KEY="..."
# export AWS_SESSION_TOKEN="..."   # if using temporary credentials
```

In production, prefer instance roles, IRSA, or IAM Identity Center over
long-lived access keys.

## 3. (Optional) Configure a model list

By default GoModel queries Bedrock's control plane and lists all on-demand
text models the account has access to. To restrict or extend the list — for
example to pin inference profile IDs or custom model ARNs — set
`BEDROCK_MODELS`:

```bash theme={null}
export BEDROCK_MODELS="anthropic.claude-3-5-haiku-20241022-v1:0,amazon.nova-lite-v1:0"
```

Or use a YAML provider block:

```yaml theme={null}
providers:
  bedrock:
    type: bedrock
    base_url: "us-east-1"
    models:
      - anthropic.claude-3-5-haiku-20241022-v1:0
      - amazon.nova-lite-v1:0
```

With the default `CONFIGURED_PROVIDER_MODELS_MODE=fallback`, the configured
list is used when Bedrock's listing call is unavailable or empty. Set
`CONFIGURED_PROVIDER_MODELS_MODE=allowlist` to expose only the configured
models and skip the upstream `ListFoundationModels` call.

## 4. Start GoModel

```bash theme={null}
go run ./cmd/gomodel
```

## 5. Verify the model registry

```bash theme={null}
curl -s http://localhost:8080/v1/models
```

Expected result:

* a `200 OK`
* Bedrock models with `owned_by` set to the originating model vendor (e.g.
  `anthropic`, `amazon`), reflecting Bedrock's `ProviderName` field.

## 6. Verify Chat Completions

```bash theme={null}
curl -s http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic.claude-3-5-haiku-20241022-v1:0",
    "messages": [{"role": "user", "content": "Reply with the single word ok."}],
    "max_tokens": 80
  }'
```

Expected result:

* a `200 OK`
* assistant content containing `ok`

Streaming (`stream: true`) and the Responses API (`/v1/responses`) work the
same way — Responses requests are bridged onto Converse internally.

## Current status

What is integrated today:

* Bedrock Runtime Converse and ConverseStream for chat and Responses
* automatic model discovery via `ListFoundationModels` (on-demand text models)
* configured model lists through `BEDROCK_MODELS` or YAML `models:` in
  fallback or allowlist mode

What is not integrated yet:

* Bedrock embeddings (the InvokeModel embedding path is model-specific and
  not yet wired up)

## Troubleshooting

* `AccessDeniedException` or `403`
  The AWS principal lacks Bedrock permissions, or model access has not been
  granted in the Bedrock console for the requested model.
* `ValidationException: ... on-demand throughput isn't supported`
  The model requires an inference profile or provisioned throughput. Set
  `BEDROCK_MODELS` (or YAML `models:`) to the inference profile ID instead of
  the base model ID.
* `model registry has no models`
  Either the region has no on-demand text models for the account, or the
  control-plane call failed. Set `BEDROCK_MODELS` to populate the registry
  from configuration, or check the IAM policy attached to the credentials.
* Wrong region inferred
  Set `BEDROCK_BASE_URL` explicitly. When passed a full endpoint URL, GoModel
  only extracts the region segment if the host ends in `.amazonaws.com`;
  custom endpoints should be paired with `AWS_REGION`.

## References

* Bedrock Runtime Converse API: [https://docs.aws.amazon.com/bedrock/latest/APIReference/API\_runtime\_Converse.html](https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_Converse.html)
* Model access in the Bedrock console: [https://docs.aws.amazon.com/bedrock/latest/userguide/model-access.html](https://docs.aws.amazon.com/bedrock/latest/userguide/model-access.html)
* AWS credential chain: [https://docs.aws.amazon.com/sdkref/latest/guide/standardized-credentials.html](https://docs.aws.amazon.com/sdkref/latest/guide/standardized-credentials.html)