localUncensoredJarvis

Like gpt4o but slower and wilder, running locally using ollama and a bunch of hugging-face models.

this readme will be improved..

Hardware

This run fine on macbook air M1, check ollama llama2 (https://github.com/ollama/ollama) to know if it can run on your device.

Install

• Make sure you have ollama installed on your machine
• Install dependencies (pip3 install -r requirements.txt)

Run

• In first terminal run ollama serve
• In second terminal run sudo python3 main.py (sudo is required on some machine)
• Ollama will download llama2-uncensored on your machine and 2 small hugging-face model (facebook/mms-tts-eng and distil-whisper/distil-medium.en).
• Add --deaf if you want to type instead of talking.
• type or say goodbye to exit.

NOTE: you may have to run ./repair.sh if the script is stuck on "waiting for subprocess", this due to the fact that a subprocess python script is needed to continuously listen to microphone, and it may get detached if program does not properly exit. (totally should be fixed i know)

Usage

• You have to say one of the following word for the microphone recording to be send to the LLM "understood", "thanks" or "do it".
• You can reset the interpreted transcription text by voice by saying "I mean", "I am saying", "reset", "I mean to say".
• You can exit by saying "goodbye", "terminate", "exit", "ciao"

Current capabilities

• Understand what you say
• Python code execution
• Using bash to interact to do anything.
• Get feedback from python/bash interpreter and fix error by itself
• Speak with a cool voice.
• Will pretty much do anything you ask.