This repository contains code for chatting with prompt-engineered AI assistants
powered by GPT through Azure API. The assistants (defined by their prompts) are
instructed to follow a set of syntax conventions, which allows backend scripts
to extract and interpret commands such as display_image(image.png). They can
show images or videos, or redirect the user to other assistants upon request.
To have a chat with an assistant in the terminal run src/console_chat.py assistant_id. The default assistant, if no argument is provided, is mental
, you need to ensure that the following file exists
../api-keys/insomnia_bot_azure.yaml where the path is relative to the project
folder. insomnia_bot_azure.yaml must have field key with the Azure API key.
To have a chatbot explain to you how the repository works, run python src/insomnia_assistant.py repository_intro (not quite finnished yet!). Running
python src/insomnia_assistant.py options shows a list of the chatbots
available. However, most of them are junk from previous testing... The prompts
that are updated and can demonstrate the abilities of the bots are hs_main,
referral, and tech_support.
By default, running python src/insomnia_assistant.py will run referral which
attempts to direct you to the appropiate bot. However, of the bots it refers to,
only tech_support is updated.
The prompts are kept in the prompts folder and its subfolders. To modify a
prompt, simply go into one of the .md files and change it. To create a new
assistant, just create a new .md file in the prompts folder (name should be
unique). The name of the file (without extension) will be interpreted as the
name of the assistant, and you can add that name as a command line argument, as
in:
python src/insomnia_assistant.py <name-of-prompt-file-you-created>
Some of the assistant are instructed to display media when it is relevant to
the conversation. See the prompt of tech_support for an example how this is
done. To show media, the prompt must instruct the bot on the path of the media
files, description of their content, and when to show them. Images and video
files are kept in the images and videos folders.
The bots can be instructed to request information as it becomes relevant, I call
this a knowledge request. See hs_main or tech_support prompts for examples
on how to instruct the bot to do this. The text files corresponding to each
knowledge request are located in the library folder. As with media files, the
bots needs to know the path and a description of their content. hs_main is a
good example where this method works quite well.
To see an example on how to instruct a bot to do a referral to another that is
better suited to answer a question, see the prompt referral.