Support for Chat Completion Endpoints with Tool Execution in Strands

[pradeepdev-1995]

Problem Statement

Enable Strands Agents to execute tools and handle function-calling workflows when using external Chat Completion APIs (e.g., https://.../chat/completions) — similar to how OpenAI’s chat/completions endpoint supports tools and tool_choice parameters.
Current Behavior
At present, Strands Agents can invoke LLMs through SDK-based models (e.g., OpenAIModel, GeminiModel, etc.), which support tool execution and structured responses via the internal orchestration framework.

However, when integrating chat-based APIs directly through HTTP calls (e.g., REST endpoints that accept model, messages, and tools in the payload), there is no current mechanism to:

Register or expose available tools to the endpoint.

Enable automatic tool selection or invocation.

Maintain consistent agent behavior between SDK-based models and REST-based chat endpoints.

Proposed Solution

Add support for executing a Strands Agent where:

The LLM backend is a generic chat completion API (not an SDK model).

The request payload includes:

model: name or identifier of the chat model.

messages: the conversation history.

tools: array of available tools/functions with schema definitions.

tool_choice: specifying auto or explicit tool invocation.

The agent runtime orchestrates:

Dynamic tool registration and schema construction.

Routing of function calls from LLM responses.

Seamless fallback between SDK and endpoint-based executions.

Use Case

curl -X POST "https://example.com/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <token>" \
  -d '{
    "model": "gpt-4.1",
    "messages": [
      {"role": "user", "content": "What’s the weather in Boston today?"}
    ],
    "tools": [
      {
        "type": "function",
        "function": {
          "name": "get_current_weather",
          "description": "Get the current weather in a given location",
          "parameters": {
            "type": "object",
            "properties": {
              "location": {"type": "string"},
              "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]}
            },
            "required": ["location"]
          }
        }
      }
    ],
    "tool_choice": "auto"
  }'

The goal is for the Strands Agent to handle the above workflow — parsing the LLM response, triggering the corresponding tool, and returning the result back to the chat completion endpoint seamlessly.

Alternatives Solutions

No response

Additional Context

No response