AWS Bedrock

Overview

Call AWS Bedrock’s model families (Claude, Nova, Mistral, Llama, Cohere, Titan) through DeepIntShield using the same OpenAI-compatible Chat Completions and Responses APIs. You select a model by ID and DeepIntShield handles the right request shape for that family. A few things are useful to know when you call Bedrock:

Model selection - the right family-specific behavior is applied automatically from the model ID you pass (e.g. bedrock/anthropic.claude-3-5-sonnet-...)
Parameters - send standard OpenAI-style fields like max_completion_tokens and stop; see Request Parameters
AWS authentication - sign in with AWS credentials (access key/secret or the standard credential chain); see AWS Authentication & Configuration
Guardrails & service tier - pass Bedrock-specific guardrail and performance settings directly in the request body
Reasoning, tools, structured output - supported with the same OpenAI-style request fields, covered in the sections below
Images - must be base64 or data-URI (remote URLs are not supported)

Model Family Support

Family	Chat	Responses	Text	Embeddings	Image Generation	Image Edit	Image Variation
Claude (Anthropic)	✅	✅	✅	❌	❌	❌	❌
Nova (Anthropic)	✅	✅	❌	❌	✅	✅	✅
Mistral	✅	✅	✅	❌	❌	❌	❌
Llama	✅	✅	❌	❌	❌	❌	❌
Cohere	✅	✅	❌	✅	❌	❌	❌
Titan	✅	✅	❌	✅	✅	✅	✅

Supported Operations

Operation	Non-Streaming	Streaming	Endpoint
Chat Completions	✅	✅	`converse`
Responses API	✅	✅	`converse`
Text Completions	✅	❌	`invoke`
Embeddings	✅	-	`invoke`
Files	✅	-	S3 (via SDK)
Batch	✅	-	`batch`
List Models	✅	-	`listFoundationModels`
Image Generation	✅	❌	`invoke`
Image Edit	✅	❌	`invoke`
Image Variation	✅	❌	`invoke`
Count Tokens	✅	-	`count-tokens`
Speech (TTS)	❌	❌	-
Transcriptions (STT)	❌	❌	-

1. Chat Completions

Request Parameters

Send standard OpenAI-compatible Chat Completions requests. The following are supported:

max_completion_tokens, temperature, top_p, stop
response_format for structured output (see Structured Output)
tools and tool_choice for function calling (see Tools)
reasoning for model thinking (see Reasoning / Thinking)
user and service_tier

Ignored Parameters

The following are not supported by Bedrock and are ignored: frequency_penalty, presence_penalty, logit_bias, logprobs, top_logprobs, seed, parallel_tool_calls.

Extra Parameters

Bedrock-specific fields can be passed directly in the request body:

curl -X POST https://app.deepintshield.com/v1/chat/completions \
  -H "Authorization: Bearer sk-bf-your-virtual-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
    "messages": [{"role": "user", "content": "Hello"}],
    "guardrailConfig": {
      "guardrailIdentifier": "guardrail-id",
      "guardrailVersion": "1",
      "trace": "enabled"
    },
    "performanceConfig": {
      "latency": "optimized"
    },
    "top_k": 40
  }'

Available Extra Parameters:

guardrailConfig - Bedrock guardrail configuration with guardrailIdentifier, guardrailVersion, trace
performanceConfig - Performance optimization with latency (“optimized” or “standard”)
additionalModelRequestFieldPaths - Pass-through for model-specific fields not in standard schema
promptVariables - Variables for prompt templates (if using prompt caching)
requestMetadata - Custom metadata for request tracking

Cache Control

Prompt caching is supported via cache control directives:

curl -X POST https://app.deepintshield.com/v1/chat/completions \
  -H "Authorization: Bearer sk-bf-your-virtual-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "This context will be cached",
            "cache_control": {"type": "ephemeral"}
          }
        ]
      }
    ],
    "system": [
      {
        "type": "text",
        "text": "You are a helpful assistant",
        "cache_control": {"type": "ephemeral"}
      }
    ]
  }'

Reasoning / Thinking

Documentation: See DeepIntShield Reasoning Reference

Reasoning/thinking support varies by model family. Use the reasoning object to enable it.

Anthropic Claude Models

reasoning.effort enables thinking
reasoning.max_tokens sets the thinking token budget
Minimum budget: 1024 tokens required; requests below this fail with error. -1 is treated as 1024.

{"reasoning": {"effort": "high", "max_tokens": 2048}}

Anthropic Nova Models

reasoning.effort sets the thinking level ("low" or "high")
reasoning.max_tokens sets the maximum reasoning tokens

{"reasoning": {"effort": "high", "max_tokens": 10000}}

Message Content

A few things to know about Bedrock message content:

Images: Only base64 / data-URI images are supported; remote image URLs are not supported.
Documents: PDF, CSV, DOC, DOCX, XLS, XLSX, HTML, TXT, and MD documents are supported as file content.
Audio: Audio input is not supported and returns an error.

Supported Chat Content Blocks

The Chat Completions request format is OpenAI-compatible. Bedrock-specific extensions (for example, a standalone cachePoint) are also accepted when using the Bedrock provider.

Block Type	Request Shape	Support
Text	`{"type":"text","text":"..."}`	✅
Image	`{"type":"image_url","image_url":{"url":"data:image/png;base64,..."}}`	✅ (base64/data URI only)
File	`{"type":"file","file":{...}}`	✅
Input audio	`{"type":"input_audio",...}`	❌ (not supported)
Standalone cache point	`{"cachePoint":{"type":"default"}}` (no outer `type` field)	✅ (Bedrock-specific extension)

Image Block Example (`image_url`)

curl -X POST https://app.deepintshield.com/v1/chat/completions \
  -H "Authorization: Bearer sk-bf-your-virtual-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
    "messages": [
      {
        "role": "user",
        "content": [
          {"type": "text", "text": "What is in this image?"},
          {
            "type": "image_url",
            "image_url": {
              "url": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAA..."
            }
          }
        ]
      }
    ]
  }'

File Block Example (`file`)

curl -X POST https://app.deepintshield.com/v1/chat/completions \
  -H "Authorization: Bearer sk-bf-your-virtual-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
    "messages": [
      {
        "role": "user",
        "content": [
          {"type": "text", "text": "Summarize this document."},
          {
            "type": "file",
            "file": {
              "file_data": "JVBERi0xLjQKJcfs...",
              "filename": "report.pdf",
              "file_type": "application/pdf"
            }
          }
        ]
      }
    ]
  }'

Note: file_data is raw base64-encoded content (no data: URI prefix, unlike image_url). Supported document formats: pdf, txt, md, html, csv, doc, docx, xls, xlsx. Use file.file_data for document payloads; file_url and file_id are not supported for Bedrock chat content.

Standalone Cache Point Example (Bedrock-specific)

{
  "role": "system",
  "content": [
    {"type": "text", "text": "Long context to cache"},
    {"cachePoint": {"type": "default"}}
  ]
}

This standalone cachePoint block is a DeepIntShield/Bedrock extension (not OpenAI-standard) and should be used only with the Bedrock provider.

Cache Control Locations

Cache directives are supported on:

System content blocks (entire system message)
User message content blocks (specific parts)
Tool definitions within tool configuration

Tools

Standard OpenAI-style tool definitions are supported. tool_choice accepts:

`tool_choice`	Behavior
`"auto"`	Model decides (default)
`"none"`	No tool calls
`"required"`	Must call a tool
Specific tool	Restricts to the named function

The function.strict field is not supported by Bedrock and is ignored. Tool call arguments are returned as a JSON string in tool_calls.

Structured Output

Bedrock has no native structured-output parameter, so DeepIntShield implements response_format for you. Send the standard response_format with a JSON schema and you receive structured JSON in the response content - no extra handling required.

{
  "response_format": {
    "type": "json_schema",
    "json_schema": {
      "name": "response",
      "schema": {
        "type": "object",
        "properties": {
          "name": {"type": "string"},
          "age": {"type": "number"}
        }
      }
    }
  }
}

The schema you provide is enforced on the model’s output, and the structured result is returned as the message content.

Response

Responses come back in the standard OpenAI-compatible shape:

finish_reason (stop, length, tool_calls)
usage.prompt_tokens / usage.completion_tokens, with cache usage rolled into prompt_tokens and broken out under prompt_tokens_details.cached_read_tokens / cached_write_tokens
reasoning_details - model thinking output (when reasoning is enabled)
Tool call arguments returned as a JSON string in tool_calls

Structured Output Response

When you request structured output, the structured JSON is returned as the message content (not as a tool call), so you consume it the same way you would from any other provider.

Streaming

Set "stream": true on Chat Completions or Responses to receive output as standard OpenAI-compatible streaming chunks (content text, tool-call arguments, and reasoning arrive incrementally).

Text Completions do not support streaming on Bedrock.

2. Responses API

Bedrock is available through the OpenAI-style Responses API on the same models as Chat Completions.

Request Parameters

Send standard OpenAI Responses requests. The following are supported:

max_output_tokens, temperature, top_p
instructions for system instructions
input as a string or array
tools and tool_choice (see Chat Completions)
reasoning for model thinking (see Reasoning / Thinking)
text for structured output

Bedrock-specific fields such as include and stop can be passed directly in the request body. Cache control is supported on instructions and input messages.

curl -X POST https://app.deepintshield.com/v1/responses \
  -H "Authorization: Bearer sk-bf-your-virtual-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
    "input": "Hello, how are you?",
    "stop": ["###"]
  }'

Response

Responses come back in the standard OpenAI Responses shape:

status (completed, incomplete)
usage.input_tokens / usage.output_tokens, with cache usage broken out under input_tokens_details.cached_read_tokens / cached_write_tokens
output items: assistant text as message, tool calls as function_call, model thinking as reasoning

Streaming

Set "stream": true to receive output as standard OpenAI Responses streaming events.

3. Text Completions (Legacy)

Send a prompt with standard parameters (max_tokens, temperature, top_p). top_k and stop can be passed directly in the request body. The response is returned in OpenAI-compatible completion shape, with text in choices[].text. Mistral models can return multiple completions.

4. Embeddings

Supported embedding models: Titan, Cohere

Request Parameters

Parameter	Notes
`input`	Text or array of texts to embed
`dimensions`	⚠️ Not supported - Titan has fixed dimensions per model
`encoding_format`	`"base64"` or `"float"`

Titan-specific: No dimension customization; fixed output size per model version.

Cohere-specific: Uses the same parameters as the standard Cohere provider.

Response

The embeddings response includes the embedding vector(s) and token usage in the standard OpenAI-compatible format.

5. Image Generation

Supported image generation models: Titan Image Generator v1, Titan Image Generator v2, Nova Canvas v1

Request Parameters

Parameter	Notes
`prompt`	Text description of the image
`n`	Number of images to generate
`negative_prompt`	What to avoid in the image
`seed`	Seed for reproducibility
`quality`	Image quality (see Quality Mapping)
`style`	Image style
`size`	Image size in `"WxH"` format (e.g., `"1024x1024"`)

Quality Mapping

The quality value maps to Bedrock’s supported quality levels:

Input Value	Result
`"low"`	Standard
`"medium"`	Standard
`"high"`	Premium
`"default"`	Standard
`"premium"`	Premium

Response

Generated images are returned in data[] as base64 (data[].b64_json).

Example Request

curl -X POST https://app.deepintshield.com/v1/images/generations \
  -H "Authorization: Bearer sk-bf-your-virtual-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bedrock/amazon.nova-canvas-v1:0",
    "prompt": "A futuristic cityscape with a flying car",
    "size": "1024x1024",
    "seed": 123,
    "negative_prompt": "bikes",
    "n": 2
  }'

6. Image Edit

Supported image edit models: Titan Image Generator v1, Titan Image Generator v2, Nova Canvas v1

Bedrock supports three image edit task types: INPAINTING, OUTPAINTING, and BACKGROUND_REMOVAL. The type field is required and must be one of these values.

Request Parameters

Parameter	Type	Required	Notes
`model`	string	✅	Model identifier (must be Titan or Nova Canvas model)
`type`	string	✅	Edit type: `"inpainting"`, `"outpainting"`, or `"background_removal"`
`prompt`	string	❌	Text description of the edit (required for inpainting/outpainting)
`image[]`	binary	✅	Image file(s) to edit (only first image used)
`mask`	binary	❌	Mask image file (for inpainting/outpainting)
`n`	int	❌	Number of images to generate (1-10, for inpainting/outpainting only)
`size`	string	❌	Image size: `"WxH"` format (e.g., `"1024x1024"`, for inpainting/outpainting only)
`quality`	string	❌	Image quality (for inpainting/outpainting only). See Quality Mapping for supported values.
`cfgScale`	float	❌	CFG scale (pass as an extra param, for inpainting/outpainting only)
`negative_text`	string	❌	Negative prompt (pass as an extra param, for inpainting/outpainting only)
`mask_prompt`	string	❌	Mask prompt (pass as an extra param, for inpainting/outpainting only)
`return_mask`	bool	❌	Return mask in response (pass as an extra param, for inpainting/outpainting only)
`outpainting_mode`	string	❌	Outpainting mode (pass as an extra param, outpainting only): `"DEFAULT"` or `"PRECISE"`

Behavior

Task Type: Set type to "inpainting", "outpainting", or "background_removal". Any other value returns an error.
Image Handling: The first image you send is used.
Task-Specific Parameters:
- Inpainting: Accepts prompt, the input image, an optional mask, and the negative_text, mask_prompt, and return_mask extra params.
- Outpainting: Accepts the same as inpainting, plus the outpainting_mode extra param ("DEFAULT" or "PRECISE").
- Background removal: Accepts only the input image; no other parameters apply.
Generation Config (inpainting and outpainting only): n (number of images), size ("WxH"), quality (see Quality Mapping), and cfgScale (extra param).

Response

Returns the same response shape as image generation: images[] (base64-encoded images) and, if return_mask was true, a base64-encoded maskImage.

Streaming: Image edit streaming is not supported by Bedrock.

7. Image Variation

Supported image variation models: Titan Image Generator v1, Titan Image Generator v2, Nova Canvas v1

Request Parameters

Parameter	Type	Required	Notes
`model`	string	✅	Model identifier (must be Titan or Nova Canvas model)
`image`	binary	✅	Image file to create variations from (supports multiple images via `image[]`)
`n`	int	❌	Number of images to generate (1-10)
`size`	string	❌	Image size: `"WxH"` format (e.g., `"1024x1024"`)
`quality`	string	❌	Image quality. See Quality Mapping for supported values.
`cfgScale`	float	❌	CFG scale (pass as an extra param)
`prompt`	string	❌	Prompt/text for variation (pass as an extra param)
`negativeText`	string	❌	Negative prompt (pass as an extra param)
`similarityStrength`	float	❌	Similarity strength (pass as an extra param): Range 0.2 to 1.0

Behavior

Image Handling: You can supply one or more input images via image[].
Variation Parameters: prompt, negativeText, and similarityStrength (range 0.2 to 1.0) can be set as extra params.
Generation Config: n (number of images), size ("WxH"), quality (extra param, see Quality Mapping), and cfgScale (extra param).

Response

Returns the same response shape as image generation: images[] (base64-encoded image variations).

Streaming: Image variation streaming is not supported by Bedrock.

8. Batch API

Request formats: requests array (CustomID + Params) or input_file_id

Pagination: Cursor-based with afterId, beforeId, limit

Endpoints:

POST /batch - Create batch
GET /batch - List batches
GET /batch/{batch_id} - Retrieve batch
POST /batch/{batch_id}/cancel - Cancel batch

Response: JSONL format with {recordId, modelOutput: {...}} or {recordId, error: {...}}

Batch statuses: Validating, InProgress, Completed, Failed, Cancelling, Cancelled, Expired

9. Files API

Upload: Multipart/form-data with file (required) and filename (optional)

Endpoints:

POST /v1/files - Upload
GET /v1/files - List (cursor pagination)
GET /v1/files/{file_id} - Retrieve metadata
DELETE /v1/files/{file_id} - Delete
GET /v1/files/{file_id}/content - Download content

Note: File purpose is always "batch", status is always "processed".

10. List Models

Request: GET /v1/models (no body)

Returns available Bedrock models with metadata. Models can be filtered by region, deployment configuration, and an allowlist (allowed_models config). When multiple keys are configured, results are aggregated across keys.

11. AWS Authentication & Configuration

DeepIntShield signs every Bedrock request with AWS Signature Version 4 (SigV4). Credentials are resolved in the following priority order, and STS AssumeRole can be layered on top of any of them.

Authentication Methods

1. Explicit Credentials

Provide access_key and secret_key directly in bedrock_key_config. Optionally include a session_token for pre-obtained temporary credentials.

{
  "bedrock_key_config": {
    "access_key": "your-aws-access-key",
    "secret_key": "your-aws-secret-key",
    "session_token": "optional-session-token",
    "region": "us-east-1"
  }
}

2. Default Credential Chain (IAM Role / Instance Profile)

Leave access_key and secret_key empty (or omit them). DeepIntShield calls AWS LoadDefaultConfig which automatically resolves credentials from the environment in this order:

Environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_SESSION_TOKEN)
EKS IRSA (AWS_WEB_IDENTITY_TOKEN_FILE + AWS_ROLE_ARN)
ECS task role
EC2 instance profile (IMDS)
~/.aws/credentials default profile

{
  "bedrock_key_config": {
    "region": "us-east-1"
  }
}

3. STS AssumeRole

Set role_arn to assume an IAM role before signing requests. AssumeRole requires a valid source identity - it works when credentials are available either via explicit access_key/secret_key in key config, or via the default credential chain (environment variables, EC2 instance profile, ECS task role, EKS IRSA, etc.). If no credentials are available from either source, AssumeRole will fail.

{
  "bedrock_key_config": {
    "role_arn": "arn:aws:iam::123456789012:role/BedrockRole",
    "external_id": "optional-external-id",
    "session_name": "my-session",
    "region": "us-east-1"
  }
}

Field	Required	Default	Notes
`role_arn`	Yes (for STS)	-	IAM role ARN to assume
`external_id`	No	-	Required when the role’s trust policy demands it
`session_name`	No	`deepintshield-session`	Identifies the session in CloudTrail logs

Setup & Configuration

How to Use ARNs and Application Inference Profiles

When using AWS Bedrock inference profiles or application inference profiles, you must split the configuration correctly to avoid UnknownOperationException:

Field	Purpose
`arn`	The ARN prefix (everything before the final `/resource-id`). Required for URL formation when using inference profiles.
`deployments`	Map logical model names to the model ID or inference profile resource ID only - not the full ARN.

Application inference profiles - use the resource ID (short alphanumeric suffix) in deployments:

{
  "bedrock_key_config": {
    "access_key": "your-aws-access-key",
    "secret_key": "your-aws-secret-key",
    "session_token": "optional-session-token",
    "region": "eu-west-1",
    "arn": "arn:aws:bedrock:eu-west-1:123456789012:application-inference-profile",
    "deployments": {
      "claude-opus-4-6": "ghi56rst",
      "claude-sonnet-4-5": "jkl78mno"
    }
  }
}

Cross-region inference profiles - use the model identifier (e.g., us.anthropic.claude-3-5-sonnet-20241022-v2:0) in deployments:

{
  "bedrock_key_config": {
    "access_key": "your-aws-access-key",
    "secret_key": "your-aws-secret-key",
    "session_token": "optional-session-token",
    "region": "us-east-1",
    "arn": "arn:aws:bedrock:us-east-1:123456789012:inference-profile",
    "deployments": {
      "claude-sonnet": "us.anthropic.claude-3-5-sonnet-20241022-v2:0"
    }
  }
}

For detailed instructions on setting up AWS Bedrock authentication including credentials, IAM roles, regions, and deployment mapping, see Provider-Specific Authentication - AWS Bedrock in the Gateway Quickstart for configuration steps in the Web UI.

Endpoints

Runtime API: bedrock-runtime.{region}.amazonaws.com/model/{path}
Control Plane: bedrock.{region}.amazonaws.com (list models)
Batch API: Via bedrock-runtime

12. Error Handling

HTTP Status Mapping:

Status	DeepIntShield Error Type	Notes
400	`invalid_request_error`	Bad request parameters
401	`authentication_error`	Invalid/expired credentials
403	`permission_denied_error`	Access denied to model/resource
404	`not_found_error`	Model or resource not found
429	`rate_limit_error`	Rate limit exceeded
500	`api_error`	Server error
529	`overloaded_error`	Service overloaded

Error Response Structure:

{
  "error": {
    "type": "invalid_request_error",
    "message": "Human-readable error message"
  }
}

Special Cases:

Cancelled requests return a request-cancelled error.
Timed-out requests return a request-timeout error.
Streaming errors are delivered in-stream with an end-of-stream indicator.
Malformed provider responses return a response-parsing error.

Caveats

Image Format Restriction

Severity: High Behavior: Only base64/data URI images supported; remote URLs not supported Impact: Requests with URL-based images fail

Minimum Reasoning Budget (Claude)

Severity: High Behavior: reasoning.max_tokens must be >= 1024 Impact: Requests with lower values fail with error

Model Family-Specific Reasoning

Severity: Medium Behavior: Reasoning/thinking support varies by model family Impact: Behavior differs for Claude vs Nova vs other families

Text Completion Streaming Not Supported

Severity: Medium Behavior: Text completion streaming returns error Impact: Streaming not available for legacy completions API

Structured Output

Severity: Low Behavior: Bedrock has no native structured-output parameter, so response_format is implemented for you under the hood Impact: None - you receive standard schema-conforming JSON in the response content

Deployment Region Prefix Handling

Severity: Low Behavior: Model IDs with region prefixes matched against deployment config Impact: Model availability depends on deployment configuration

AWS Bedrock

Overview

Model Family Support

Supported Operations

1. Chat Completions

Request Parameters

Ignored Parameters

Extra Parameters

Cache Control

Reasoning / Thinking

Anthropic Claude Models

Anthropic Nova Models

Message Content

Supported Chat Content Blocks

Image Block Example (image_url)

File Block Example (file)

Standalone Cache Point Example (Bedrock-specific)

Cache Control Locations

Tools

Structured Output

Response

Structured Output Response

Streaming

2. Responses API

Request Parameters

Response

Streaming

3. Text Completions (Legacy)

4. Embeddings

Request Parameters

Response

5. Image Generation

Request Parameters

Quality Mapping

Response

Example Request

6. Image Edit

7. Image Variation

8. Batch API

9. Files API

10. List Models

11. AWS Authentication & Configuration

Authentication Methods

1. Explicit Credentials

2. Default Credential Chain (IAM Role / Instance Profile)

3. STS AssumeRole

Setup & Configuration

How to Use ARNs and Application Inference Profiles

Endpoints

12. Error Handling

Caveats

Image Block Example (`image_url`)

File Block Example (`file`)