Feature Request: Support for Multimodal Input Format in langchain-aws

[KwangGan]

Hello,
I would like to request a feature enhancement for langchain-aws to support multimodal input format. Currently, I can work with the existing functionality without issues, but adding multimodal support would allow for more streamlined control and easier code maintenance in my project.
For reference, I'm following the implementation approach described in the LangChain documentation: https://python.langchain.com/docs/how_to/multimodal_inputs/

Thank you for your consistent maintenance of this project.

[michaelnchin]

Hi @KwangGan, thank you for submitting an issue.

Multimodal content in the LangChain standard format is supported as of v0.2.21. (See: #455)

Here is a code sample that adapts the documented base64 PDF example to use ChatBedrockConverse. Note that the extra name field is a requirement specific to Bedrock Converse API.

import base64
import httpx
from langchain_core.messages import HumanMessage

pdf_url = "https://pdfobject.com/pdf/sample.pdf"
pdf_data = base64.b64encode(httpx.get(pdf_url).content).decode("utf-8")

model = ChatBedrockConverse(
    model_id="us.anthropic.claude-3-7-sonnet-20250219-v1:0",
    region_name="us-west-2"
)

message = {
    "role": "user",
    "content": [
        {
            "type": "text",
            "text": "Describe the document:",
        },
        {
            "name": "a_pdf_input",
            "type": "file",
            "source_type": "base64",
            "data": pdf_data,
            "mime_type": "application/pdf",
        },
    ],
}
response = llm.invoke([message])
print(response.text())

Let us know if this works for you.

[michaelnchin]

Closing due to >1 month of inactivity - please re-open with any additional information.