Skip to content

Latest commit

 

History

History
39 lines (32 loc) · 2.06 KB

README.md

File metadata and controls

39 lines (32 loc) · 2.06 KB
Raw

jarvis_ai

Conversational AI Program similar to Jarvis from MCU

Developed By: Greg Salmon

Installation

  1. Download the zip file
  2. Install necessary packages using the terminal command pip install -r requirements.txt
  3. In terminal, Create a .env file using the command touch .env
    • Inside the .env file, type: OPENAI_API_KEY="Your Key Here"

Running the Program

This program will run continuously until ended

  • Run the command sudo python3 jarvis.py

    • Important to note: We use the sudo attribute because the program utilizes the keyboard library, which essentially allows the program to view keystrokes.
    • When you first run the program, it may fail and prompt you to update your computer accessibility settings.
    • You may also be prompted to allow microphone access
    • There is no malicious code to worry about but I recommend always previewing the code prior to running it on your computer if you are concerned

  • To record your message to Jarvis, press "esc" when prompted

  • To end your recording, press "esc" again

  • Jarvis will provide a response within 15 seconds, delay depends on length of question and memory

  • To quit the program, press "ctrl+C"

  • Message history is stored in the /conversation_history/ directory

    • To view an example message history, see /conversation_history/example_message/example_message.csv

Comments

  • This program currently runs via the openai API, which costs money.
    • Applications used include Whisper, Chat Generation, and Text to Speech
  • BE WARNED, while it is inexpensive, accidentally leaving the recording running will cause the openai billing to significantly increase, as the audio models cost the most

Final Thoughts

I originally developed this program with open source tools found on HuggingFace. I decided to transition to openai due to its speed. My previous iteration took around 30-45 seconds per response, which is not acceptable for a real time chatbot.

Future plans include:

  • Moving towards all open-source/free models
  • Cleaning code
  • Decrease time to response