An extra-minimalistic CLI to interact with ChatGPT via OpenAI API.
I created this CLI tool because I use ChatGPT a lot in my daily tasks for a variety of purposes including coding, copywriting, brainstorming and "googling". I was looking for a more immediate way to access it, and as I spend lots of time on the CLI, the most usable way for me was to have it built into the CLI itself.
Despite being only 60 lines of code, ChatGPT-CLI allows to:
- interact with ChatGPT with a chat-like experience
- remember conversations using memory
- store conversations locally
- load and continue conversations
- set the temperature of the model
- limit the memory depth to save tokens and contain costs
The tool runs as a Python script, so you only need to:
- download the repo
- create the
.secretsandhistoryfolders - install dependencies
pip install typer
pip install openai
And you are good to invoke it with python3 app.py.
For usability, I like to add an alias to my Mac and invoke it with one command:
echo "alias gptcli='python3 path-to/app.py'" >> ~/.zshrc && source ~/.zshrc
If you wanna run it as an alias, is a good idea to customize these two environment variables to include the full path of the script.
API_KEY_FILE = "path-to/.secrets/openai_key.txt"
HISTORY_PATH = "path-to/history/"
python3 app.py Start a new chat.
python3 app.py --model=gpt-4 Uses GPT-4 as LLM.
python3 app.py --model=gpt-4 -ml=4 Uses GPT-4 as LLM and limit the memory to the last 4 messages.
python3 app.py -t=0.9 Set the temperature to 0.9 (more creative).
python3 app.py --load=history/marketing_plan.txt Load an existing conversation into memory.
python3 app.py --help to get fresh info about parameters and usage.
To save a conversation, type save as User prompt.
To exit, type exit as User prompt.
- improve the setup process
- save on the same file (when loaded from conversation)
- progress bar for a better experience on long LLM queries
Just fork the repo!