About the method for multimodal requests

[tbrand]

I want to execute multimodal requests with Strands Agents (v0.2.1).
The stream_async function always requires a prompt: str parameter when executed (source) and it gets added to messages as a user message. (relevant section) With this specification, it's not possible to handle use cases where you want to send images or files and have their content analyzed. Is there a way to omit the prompt? Note that the response needs to be streaming.

[tbrand]

As a workaround, we are currently implementing multimodal requests by specifying "." in the prompt as shown below. An empty string results in a validation error.

agent = Agent(
    system_prompt=system_prompt,
    model=model,
    tools=tools,
    messages=messages, # This message contains text content and image content.
)

async for event in agent.stream_async('.'):
    pass

[zastrowm]

#212 should address some of these use cases - the intent is to allow different types of input other than text to be passed into stream_async and __call__.