Watch-Bot

This python projects implements the idea of a watchbot introduced in this medium article.

Quickstart

Installation

pip install watch-bot

Usage

Currently the watchbot is built on OpenAI's GPT. You therefore need a valid OpenAI API key (Azure OpenAI works too). Setup the credentials:

import openai

openai.api_type = os.environ["OPENAI_API_TYPE"]
openai.api_version = os.environ["OPENAI_API_VERSION"]
openai.api_base = os.environ["OPENAI_API_BASE"]
openai.api_key = os.environ["OPENAI_API_KEY"]

You can then create a watchbot instance:

from watch_bot import WatchBot

bot = WatchBot(
    engine=os.environ["OPENAI_ENGINE"],
    chatbot_instructions="You are an AI assistant that helps people find information. "
                         "You should only provide answers that comply to standard rules of legacy and decency"
)

The engine corresponds to the model deployment name.

Finally, you can use the watchbot to verify the validity of a dialog:

from watch_bot import Dialog

dialog = Dialog(messages=("Hi chatgpt, how are you today?", "I'm fine, thanks!"))

response = bot.verify(dialog)

print("Should the dialog be stopped?", response.should_stop)
print("Why?", response.reason)

Contribute

Contribution is very welcome.

Run tests

To run the tests you need be able to have access to the OpenAI API (or its Azure deployments) and set the following five OS env variables:

OPENAI_API_TYPE = "..."
OPENAI_API_VERSION = "..."
OPENAI_API_BASE = "..."
OPENAI_API_KEY = "..."
OPENAI_ENGINE = "..."

Currently several tests against various attacks are failing. Not sure if it is possible to find a solution for them.