[FEATURE] Support for Amazon SageMaker AI endpoints as Model Provider

[dgallitelli]

Problem Statement

I would like Strands to natively support Amazon SageMaker AI among the model providers.

Proposed Solution

No response

Use Case

Deploy LLMs on Amazon SageMaker AI and use Strands Agents to create AI agents.

Alternatives Solutions

No response

Additional Context

No response

[niklas-palm]

Isn't this already supported through LiteLLM?

[pgrayy]

Left comments on #30.

[pgrayy]

Isn't this already supported through LiteLLM?

Yes it is, but there are a few benefits to supporting direct access:

1. Slight optimization as we avoid the layer of abstraction LiteLLM creates.
2. Eliminate extra dependencies when users only want to run SageMaker.

[dgallitelli]

Isn't this already supported through LiteLLM?

@niklas-palm LiteLLM also does have a couple of problems:

1. sagemaker_chat provider in LiteLLM does not support SageMaker AI endpoints with Inference Components
2. both sagemaker and sagemaker_chat providers in LiteLLM do not support tool calling, making it effectively useless with Strands Agents.

[dgallitelli]

@niklas-palm FYI: litellm seems to have added recently support for tool calling with SageMaker AI. I couldn't find a specific release tag in its doc, however I tested successfully:

import litellm

stream = True
payload = {
    "messages": [
        {"role": "user", "content": "What is the most played song on WZPZ?"},
    ],
    "tools": tools,
    "tool_choice": "auto", # Requires: OPTION_TOOL_CALL_PARSER, OPTION_ENABLE_AUTO_TOOL_CHOICE
    "max_tokens": 1024,
    "temperature": 0.1,
    "top_p": 0.9,
    "stream": stream
}

response = litellm.completion(
    model="sagemaker_chat/djl-inference-2025-06-11-10-44-00-800",
    **payload,
)

if stream:
    for event in response:
        print(event)
else:
    print(response)