Python: Support File Uploads in DevUI for Document Workflows

[ShivamGoyal03]

Feature Request: Support File Uploads in DevUI for Document Workflows

I am using PrivyDoc, a workflow built with MAF + Foundry Local, to process documents like PDFs and Word files. The workflow extracts text and performs analysis.

When using DevUI, I can see the workflow running, but it only allows typing text. There is no option to upload files, which prevents testing workflows like PrivyDoc fully in DevUI.

Current Behavior

• DevUI inspects the starting executor input type.
• If the input type is str, DevUI shows a textbox.
• File inputs cannot be uploaded because DevUI does not detect complex input types like documents or files.

Problem

• Many workflows need uploading PDFs, Word files, or other structured inputs.
• Right now, DevUI does not support these input types, limiting testing and the developer experience.
• Checking MIME type alone is not enough because the executor input type must be set beforehand. MIME type only helps after the file enters the workflow.

Proposed Approach / Use Case

• Allow DevUI to recognize a File or Document type input.
• Show a file upload UI when the workflow expects a document.
• Within the workflow, MIME type can be used to:
  • Decide which parser or extraction method to use (PDF, Word, text)
  • Route files to specialized sub-agents for processing
  • Validate file types before processing
  • Handle large files with metadata-based methods such as chunking
• Ensure typed message routing still functions correctly.

Use Case

• A user uploads a PDF or Word document to start a workflow.
• DevUI should permit the upload and pass the document correctly to the workflow.
• Then, the workflow processes the document based on its MIME type, extracts content, and executes the intended analysis.

Benefit

• Makes DevUI more adaptable for real-world workflows that handle documents, images, or other structured data.
• Improves developer testing and eases the creation of document-processing workflows.

Cc: @victordibia @moonbox3

[victordibia]

@ShivamGoyal03

Can you show a minimal code-based version of what you a simple workflow in code like this would be? Maybe something like this?

import logging
import os
from typing import Optional

from pydantic import BaseModel
from agent_framework import Executor, WorkflowBuilder, WorkflowContext, handler
from agent_framework.azure import AzureOpenAIChatClient
from agent_framework.devui import serve
from typing_extensions import Never


# Define a Pydantic model for the document input
class DocumentInput(BaseModel):
    file_name: str
    file_content: bytes  # DevUI can accept file uploads for byte fields
    mime_type: Optional[str] = None


class DocumentProcessor(Executor):
    """Executor to process an uploaded document."""

    @handler
    async def run(
        self, 
        input: DocumentInput, 
        ctx: WorkflowContext[DocumentInput]
    ) -> None:
        # Example processing: just report file info and content length
        file_info = (
            f"Received file '{input.file_name}' "
            f"of type '{input.mime_type or 'unknown'}' with size {len(input.file_content)} bytes."
        )
        # Yield output back to the workflow
        await ctx.yield_output(file_info)


def main():
    logging.basicConfig(level=logging.INFO)
    logger = logging.getLogger(__name__)

    chat_client = AzureOpenAIChatClient(
        api_key=os.environ.get("AZURE_OPENAI_API_KEY"),
        azure_endpoint=os.environ.get("AZURE_OPENAI_ENDPOINT"),
        api_version=os.environ.get("AZURE_OPENAI_API_VERSION", "2024-10-21"),
        model_id=os.environ.get("AZURE_OPENAI_CHAT_DEPLOYMENT_NAME", "gpt-4o"),
    )

    processor = DocumentProcessor(id="document_processor")

    workflow = (
        WorkflowBuilder(
            name="Document Upload Workflow",
            description="Workflow that accepts a document upload and processes it.",
        )
        .set_start_executor(processor)
        .build()
    )

    logger.info("Starting DevUI for document upload workflow at http://localhost:8090")
    serve(entities=[workflow], port=8090, auto_open=True)


if __name__ == "__main__":
    main()

[ShivamGoyal03]

Hi @victordibia ,

Thanks for the quick follow-up! Here’s the project I mentioned earlier: PrivyDoc

In this workflow, I’m using Chainlit as the interface, where users can upload attachments (PDFs or Word documents), and the app processes the content after extraction.

I tried initializing and running the same flow through DevUI, but since it only supports text input right now, I couldn’t replicate the document upload behavior. That’s the main gap I’m running into - there’s no option in DevUI to upload files, so workflows like PrivyDoc can’t be tested in full.

The example you shared matches my use case well, where a DocumentInput model (with bytes and MIME type) could allow DevUI to accept file uploads.