Promptflow Copilot

An intelligent copilot designed for promptflow

Objective

The Promptflow Copilot is an intelligent assistant built on the ChatGPT platform, designed to interactively assist Promptflow users with:

• Automatically generating promptflow to accomplish the given goal [Done]
• Understanding users' pre-existing code and converting it into promptflow [Done]
• Assisting users in generating bulk test input data [Done]
• Assisting users in creating an evaluation flow [Done]
• Comprehending existing promptflow [Done]
• Assisting users in adding variations for llm node [Done]

How to use

• From root folder, install the requirements using

  pip install -r requirements.txt

• Create a pfcopilot.env file in root folder, reference pfcopilot.env.example to set the corresponding environment variables in your own pfcopilot.env file.

Promptflow Copilot relys on the function calling features of OpenAI API. You must use models or deployments that support function calling. learn more about function calling

• Currently, we provide two ways to chat with promptflow copilot

  • Chat with UI: from the root folder, run

    python main.py

    Then you can chat with it like that:

  You can start a new chat by click the New chat button.

  • Chat with CLI: from the root folder, run

    python copilot_cli.py

    Then you can chat with it like that:

  You can end the chat by type exit in the command line, or start a new chat by type new chat in the command line.

• chat with promptflow copilot

for example:

1. Ask copilot to create a promptflow to achieve your goal:

   my goal: check if there are gramma mistakes in a github repo's files, the file may be written with python, c, go or any programming language; if found gramma mistakes, create a pull request to fix the gramma mistakes

2. Ask copilot to create promptflow based on your own python application

   I have a python program in my_app.py, please convert it into a flow

Your app.py can be like this:

search_query = ''
bing_engine = BingEngine()
google_engine = GoogleEngine()
bing_results = bing_engine.search(search_query)
google_results = google_engine.search(search_query)

men_charactor = "You are a man. compare the search results from bing and google, and choose the one you like."
women_charactor = "You are a woman. compare the search results from bing and google, and choose the one you like."

men_choice = choose(men_charactor, bing_results, google_results)
women_choice = choose(men_charactor, bing_results, google_results)

return {'men': men_choice, 'women': women_choice}

or

I have a python program in the folder C:\LangchainTests\chat_with_pdf, can you understand it and help to convert it into a flow

3. Ask copilot to generate bulktest input data

   geneate bulktest inputs data for the flow

4. Ask copilot to generate evaluation flow

   generate evaluation flow for the flow

5. Ask copilot to understand an existing promptflow

   I have a promptflow in the folder C:\Tests\myflow, can you understand it

6. Ask copilot to add variant for llm node

   Please add variant for my_llm node in my flow

How does it work

Known issue

• Sometimes, pfcopilot may not automatically dump the generated flow into local folder. If that happens, please tell pfcopilot to do that explicitlly.
• Sometimes, the generated flow may lack some python files when the flow is converted from local file(s). Please give pfcopilot furthur instructions to let it generate the missing files for you.