Bedrock Converse¶

AWS Bedrock LLM provider using the Converse API. Supports streaming, tool use, and all Bedrock-hosted models.

Configuration¶

Server-Side (Gateway Config)¶

providers:
  llm:
    backend: "agentic_primitives_gateway.primitives.llm.bedrock.BedrockConverseProvider"
    config:
      region: "us-east-1"
      default_model: "us.anthropic.claude-sonnet-4-20250514-v1:0"

Parameter	Default	Description
`region`	`us-east-1`	AWS region for Bedrock
`default_model`	`us.anthropic.claude-sonnet-4-20250514-v1:0`	Model used when the client doesn't specify one

Environment Variables¶

Variable	Description
`AWS_REGION`	AWS region (fallback if not in config)
`AWS_ACCESS_KEY_ID`	AWS access key
`AWS_SECRET_ACCESS_KEY`	AWS secret key
`AWS_SESSION_TOKEN`	Session token (for temporary credentials)

Using the LLM API¶

Non-Streaming¶

curl -X POST http://localhost:8000/api/v1/llm/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "us.anthropic.claude-sonnet-4-20250514-v1:0",
    "messages": [{"role": "user", "content": "What is 2+2?"}],
    "temperature": 0.7
  }'

Streaming¶

curl -X POST http://localhost:8000/api/v1/llm/completions/stream \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [{"role": "user", "content": "Explain quantum computing"}]
  }'

Returns SSE events: content_delta, tool_use_start, tool_use_delta, tool_use_complete, message_stop, metadata.

With Tool Use¶

curl -X POST http://localhost:8000/api/v1/llm/completions \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [{"role": "user", "content": "What is the weather in Seattle?"}],
    "tools": [
      {
        "name": "get_weather",
        "description": "Get current weather for a city",
        "input_schema": {"type": "object", "properties": {"city": {"type": "string"}}}
      }
    ],
    "tool_choice": "auto"
  }'

Using with the Python Client¶

LLM Gateway Model (Strands)¶

from agentic_primitives_gateway_client import AgenticPlatformClient

client = AgenticPlatformClient("http://localhost:8000", aws_from_environment=True)

# Model routed through the gateway (uses Bedrock behind the scenes)
model = client.get_model(format="strands")

LLM Gateway Model (LangChain)¶

model = client.get_model(format="langchain")

Direct Completions¶

result = await client.completions({
    "model": "us.anthropic.claude-sonnet-4-20250514-v1:0",
    "messages": [{"role": "user", "content": "Hello"}],
})

How It Works¶

The provider translates between the gateway's internal message format and the Bedrock Converse API:

Messages: gateway flat format (role + content + tool_calls) is converted to Bedrock content blocks (text, toolUse, toolResult)
System prompts: extracted from messages and passed as the top-level system parameter
Streaming: uses converse_stream() with a background thread draining boto3's sync iterator into an asyncio.Queue for async iteration
Tool calls: tool definitions are converted to Bedrock toolConfig format; tool results are converted back

Per-request AWS credentials are resolved from headers (X-AWS-Access-Key-Id, etc.) or the environment, ensuring each request uses the caller's permissions.

Running Multiple Regions¶

providers:
  llm:
    default: "bedrock-us"
    backends:
      bedrock-us:
        backend: "agentic_primitives_gateway.primitives.llm.bedrock.BedrockConverseProvider"
        config:
          region: "us-east-1"
          default_model: "us.anthropic.claude-sonnet-4-20250514-v1:0"
      bedrock-eu:
        backend: "agentic_primitives_gateway.primitives.llm.bedrock.BedrockConverseProvider"
        config:
          region: "eu-west-1"
          default_model: "eu.anthropic.claude-sonnet-4-20250514-v1:0"

Prerequisites¶

pip install agentic-primitives-gateway[agentcore]
AWS credentials with Bedrock model access
Model access enabled in the Bedrock console for the target region