microsoft/promptflow — reverse-engineered prompt

Build me a Python tool called Prompt flow that helps people create and improve LLM apps from prototype to production. I want it to let me make simple executable flows that connect prompts, model calls, Python code, and other tools, then run those flows locally, inspect what happened, and iterate quickly when something looks wrong.

Please include a good command line experience so I can spin up a chatbot example from a template, add an OpenAI or Azure OpenAI connection with my API key, and chat with the flow interactively. I also want a way to test flows against larger datasets, compare quality and performance, and use that as part of a repeatable evaluation process before deployment.

Make it feel practical and usable, with clear docs, examples, and a basic path from development to deployment and monitoring. If there are existing docs or references online, feel free to check them so the commands and behavior match current expectations.