Python: How to provide input to workflow in devui?

[pamelafox]

I'm running a workflow based off the writer/reviewer sample and visualizing it in devui, but I can't figure out how to provide input to the workflow, the equivalent of workflow.run_stream(USER_INPUT). Is it possible?

[moonbox3]

Hi @pamelafox, once DevUI loads, and you see the workflow, do you see the "Configure & Run" button near the top right?

I can then populate the initial string input (based on the type that the first Executor expects) and run the workflow:

Producing the result:

It looks like your workflow completed, you didn't provide any input to kick it off?

[pamelafox]

This is all I see:

With this code:

# Copyright (c) Microsoft. All rights reserved.

import asyncio
import os

from agent_framework import AgentRunUpdateEvent, WorkflowBuilder, WorkflowOutputEvent
from agent_framework.azure import AzureOpenAIChatClient
from agent_framework.devui import serve
from agent_framework.openai import OpenAIChatClient
from azure.identity import AzureCliCredential, DefaultAzureCredential
from dotenv import load_dotenv
from pydantic import Field
from rich import print
from rich.logging import RichHandler
from agent_framework.devui import serve


"""
Sample: Agents in a workflow with streaming

A Writer agent generates content, then a Reviewer agent critiques it.
The workflow uses streaming so you can observe incremental AgentRunUpdateEvent chunks as each agent produces tokens.

Purpose:
Show how to wire chat agents directly into a WorkflowBuilder pipeline where agents are auto wrapped as executors.

Demonstrate:
- Automatic streaming of agent deltas via AgentRunUpdateEvent.
- A simple console aggregator that groups updates by executor id and prints them as they arrive.
- The workflow completes when idle and outputs are available in events.get_outputs().

Prerequisites:
- Azure OpenAI configured for AzureOpenAIChatClient with required environment variables.
- Authentication via azure-identity. Use AzureCliCredential and run az login before executing the sample.
- Basic familiarity with WorkflowBuilder, edges, events, and streaming runs.
"""

load_dotenv(override=True)

"""Build and run a simple two node agent workflow: Writer then Reviewer."""
# Create the Azure chat client. AzureCliCredential uses your current az login.
chat_client = AzureOpenAIChatClient(
    credential=DefaultAzureCredential(),
    deployment_name=os.environ.get("AZURE_OPENAI_CHAT_DEPLOYMENT"),
    endpoint=os.environ.get("AZURE_OPENAI_ENDPOINT"),
    api_version=os.environ.get("AZURE_OPENAI_VERSION"),
)
# Define two domain specific chat agents. The builder will wrap these as executors.
writer_agent = chat_client.create_agent(
    instructions=(
        "You are an excellent content writer. You create new content and edit contents based on the feedback."
    ),
    name="writer_agent",
)

reviewer_agent = chat_client.create_agent(
    instructions=(
        "You are an excellent content reviewer."
        "Provide actionable feedback to the writer about the provided content."
        "Provide the feedback in the most concise manner possible."
    ),
    name="reviewer_agent",
)

# Build the workflow using the fluent builder.
# Set the start node and connect an edge from writer to reviewer.
workflow = WorkflowBuilder().set_start_executor(writer_agent).add_edge(writer_agent, reviewer_agent).build()
serve(entities=[workflow], auto_open=True, tracing_enabled=True)

[pamelafox]

And yeah, it did complete, as it considered the first input to be an empty string, so it kind of worked.

[moonbox3]

Ah, interesting, thanks for providing the sample code. We wrap the agent instances as an AgentExecutor. Looks like there needs to be some translation to allow DevUI to allow for input. Looks like in whatever introspection is occurring, something is getting missed. We'll work on getting this updated.

[moonbox3]

Attached PR has a fix: