DIAL Bedrock Adapter

warning

The standalone DIAL OpenAI, Bedrock, and Vertex AI adapters will be deprecated soon. Use DIAL Model Provider instead: it combines these three adapters in one service, and serves this adapter under /compat/bedrock. To move your models, see Migrate from the standalone adapters.

The DIAL Bedrock Adapter connects DIAL Core to the chat, image, and embedding models of Amazon Bedrock: Anthropic Claude, Amazon Nova and Titan, Meta Llama, Mistral, DeepSeek, Stability AI, Cohere, and others. It also reaches Claude through the Anthropic API, serves the native Anthropic Messages API for Claude Code, and can translate Anthropic Messages requests for models that speak only Chat Completions. This page is for platform engineers who connect Bedrock models to DIAL.

info

• Repository
• Releases
• Docker Hub image tags

How it works

• A fixed list of models. The adapter serves the models it knows, by their Bedrock model IDs, such as anthropic.claude-sonnet-4-5-20250929-v1:0. Any other ID answers 404, until you map it to a supported model with compatibility mode. See Supported models.
• Three implementations. Depending on the model, the adapter calls the Bedrock Converse API, the original Bedrock API, or the Anthropic SDK, which gives finer control over Claude.
• Credentials. AWS credentials and the region come from the adapter's environment variables, or per upstream from extraData in DIAL Core, so one model can span several AWS accounts and regions.

DIAL APIs it serves

API	Adapter endpoint
Chat Completions	POST /openai/deployments/<model id>/chat/completions
Embeddings	POST /openai/deployments/<model id>/embeddings
Tokenize	POST /openai/deployments/<model id>/tokenize
Truncate prompt	POST /openai/deployments/<model id>/truncate_prompt
Configuration	GET /openai/deployments/<model id>/configuration
Anthropic Messages passthrough	/anthropic/v1/messages and the other Claude API endpoints Bedrock implements
Anthropic Messages to Chat Completions translator	POST /openaiChatCompletions/anthropic/v1/messages

Which models support /tokenize, /truncate_prompt, tools, and /configuration is in Supported models.

Base configuration

From ai-dial-samples. The adapter service in Docker Compose, with AWS credentials from the environment:

adapter-dial-bedrock:
  image: epam/ai-dial-adapter-bedrock:0.45.0
  environment:
    DIAL_URL: http://core:8080
    LOG_LEVEL: INFO
    AWS_DEFAULT_REGION: us-east-1
    AWS_ACCESS_KEY_ID: <AWS_ACCESS_KEY_ID>
    AWS_SECRET_ACCESS_KEY: <AWS_SECRET_ACCESS_KEY>

A Claude model in the DIAL Core configuration. The adapter URL carries the Bedrock model ID, and the model needs no upstream URL:

{
  "models": {
    "claude-sonnet-4": {
      "type": "chat",
      "displayName": "Claude on Bedrock",
      "endpoint": "http://adapter-dial-bedrock:5000/openai/deployments/anthropic.claude-sonnet-4-20250514-v1:0/chat/completions",
      "features": { "toolsSupported": true },
      "defaults": { "max_tokens": 4096 }
    }
  }
}

Claude needs max_tokens; set it per model in defaults rather than through CLAUDE_DEFAULT_MAX_TOKENS, and keep it within the model's maximum output.

Deployment

The adapter ships as the epam/ai-dial-adapter-bedrock image. Deploy it with the DIAL Helm charts, or in Docker Compose as above. Grant its AWS identity access to Bedrock; for exact token counting of Claude on the mantle client, it also needs the bedrock-mantle:CountTokens permission. To deploy a model in your AWS account, see Bedrock model deployment.