AWS Bedrock

Configuration

credentials:
  - name: "bedrock_us_east"
    type: "bedrock"
    api_key: "os.environ/AWS_BEDROCK_API_KEY"
    base_url: "https://bedrock-runtime.us-east-1.amazonaws.com"
    rpm: 60
    tpm: 100000

Required Fields

Field	Description
`api_key`	AWS Bedrock API key / token (supports `os.environ/VAR_NAME`)
`base_url`	Bedrock Runtime endpoint URL (region-specific)

The api_key is sent as Authorization: Bearer <api_key> to the Bedrock Runtime API.

Common Base URLs by Region

Region	Base URL
us-east-1	`https://bedrock-runtime.us-east-1.amazonaws.com`
us-west-2	`https://bedrock-runtime.us-west-2.amazonaws.com`
eu-central-1	`https://bedrock-runtime.eu-central-1.amazonaws.com`
ap-northeast-1	`https://bedrock-runtime.ap-northeast-1.amazonaws.com`

Model IDs

Use AWS Bedrock model IDs in the model field. Cross-region inference profiles are supported:

us.anthropic.claude-sonnet-4-20250514-v1:0
us.anthropic.claude-3-5-sonnet-20241022-v2:0
us.anthropic.claude-3-haiku-20240307-v1:0
anthropic.claude-3-5-haiku-20241022-v1:0

OpenAI-Compatible API

The router accepts requests in OpenAI Chat Completion format and automatically converts them to AWS Bedrock InvokeModel format. The conversion pipeline:

OpenAI → Anthropic Messages API format (full parameter conversion)
Bedrock-specific adjustments: remove model (embedded in URL), remove stream, add anthropic_version: "bedrock-2023-05-31"
Response: Bedrock (Anthropic) → OpenAI format

Request URL Construction

Mode	URL
Non-streaming	`{base_url}/model/{model_id}/invoke`
Streaming	`{base_url}/model/{model_id}/invoke-with-response-stream`

Supported Parameters

Same as the Anthropic provider:

OpenAI Parameter	Notes
`temperature`	Set to 1.0 automatically when thinking is enabled
`top_p`
`max_tokens`	Defaults to 4096 if not set
`max_completion_tokens`	Fallback if `max_tokens` is empty
`stop`	Accepts string or array
`stream`
`tools`	Full conversion (see Tool Calling)
`tool_choice`
`user`	Mapped to `metadata.user_id`
`reasoning_effort`	See Extended Thinking

extra_body Parameters

Parameter	Description
`extra_body.top_k`	Top-K sampling
`extra_body.thinking`	Direct thinking config (`{"type": "enabled", "budget_tokens": 15000}`)

Unsupported Parameters

n, frequency_penalty, presence_penalty, seed, response_format, logprobs, top_logprobs, modalities, service_tier, store, parallel_tool_calls, prediction

Embeddings and image generation are not supported by this provider.

Message Conversion

Same as Anthropic provider.

Tool Calling

Same as Anthropic provider. Supported tool types:

OpenAI Tool Type	Anthropic (Bedrock) Type
`function`	Standard function tool
`computer_use`	`computer_20241022`
`text_editor`	`text_editor_20241022`
`bash`	`bash_20241022`
`web_search` / `web_search_preview`	`web_search_20250305`

Extended Thinking

Supported for reasoning-capable models (e.g. claude-sonnet-4). Same interface as Anthropic provider.

response = client.chat.completions.create(
    model="us.anthropic.claude-sonnet-4-20250514-v1:0",
    messages=[{"role": "user", "content": "Solve this step by step..."}],
    reasoning_effort="high",  # minimal, low, medium, high
)

reasoning_effort	budget_tokens
`minimal`	1,000
`low`	5,000
`medium`	15,000
`high`	30,000
`none` / `disable`	Disabled

When thinking is enabled, temperature is automatically set to 1.0.

Thinking content is returned in the reasoning_content field of the response message.

Content Types

Content Type	Support	Notes
Text	Full	String or `{"type": "text"}` blocks
Image (base64)	Full	`data:image/...;base64,...` → Anthropic base64 source
Image (URL)	Full	HTTP/HTTPS URLs → Anthropic URL source
Document (base64)	Full	`application/` and `text/` MIME types only
Audio	Placeholder	Replaced with `[Audio input: <format> format - not supported by Anthropic API]`
Video	Placeholder	Replaced with `[Video: <url>]`

Streaming

AWS Bedrock uses a binary Event Stream framing protocol instead of SSE. The router decodes it transparently:

Binary AWS Event Stream frames decoded on the fly
Each frame payload (base64-encoded Anthropic event JSON) converted to SSE
Anthropic SSE events converted to OpenAI streaming format

The client receives standard OpenAI-compatible SSE:

stream = client.chat.completions.create(
    model="us.anthropic.claude-3-5-sonnet-20241022-v2:0",
    messages=[{"role": "user", "content": "Hello"}],
    stream=True,
)

for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

Finish Reasons

Bedrock/Anthropic stop_reason	OpenAI finish_reason
`end_turn`	`stop`
`max_tokens`	`length`
`tool_use`	`tool_calls`
`stop_sequence`	`stop`

Token Counting

Anthropic Field	OpenAI Mapping
`input_tokens`	`prompt_tokens`
`output_tokens`	`completion_tokens`
`cache_read_input_tokens`	`prompt_tokens_details.cached_tokens`
`cache_creation_input_tokens`	Tracked internally for billing