ARCHIVED

This is no longer maintained because Simon Willison has such a great tool at https://github.com/simonw/llm

ASKGPTCLI

A command line interface to ChatGPT, for when you have questions and don't want to switch to Chrome.

Install

Either clone the repository or download the code as a zip and save it to a directory of your choice.

Install python. On macOS you can use homebrew

brew install python
source myenv/bin/activate

python pip install -r requirements.txt

Next, edit your .bashrc or .zshrc file to include your OpenAI API key:

export OPENAI_API_KEY="XXX"

Exit the terminal and reopen to the same directory.

You can use it right from the command line in this directory. If you want it to be available everywhere then add this alias to your .bashrc or .zshrc file:

GPTDIR="~/path/to/where/you/installed/this"
alias askgpt='source $GPTDIR/env/bin/activate && python3 $GPTDIR/askgpt.py "$@" && deactivate'

Usage

If you added the alias, then just use askgpt from your terminal.

There are three things you need to know about input

• Input does not enter until you enter two blank lines
• If you start code with ``` then it will accept input until you enter ``` and two blank lines
• exit or quit will gracefully close the program

Reset the chat

Sometimes GPT gets confused with too much background information. You can reset the chat by typing reset chat You can name your new chat with reset chat=myName, although that change only matters to the SQLite file.

This effectively lets you switch between chats if you want as well.

See Chat History

/history will output the entire history of the currently named chat. This may include entries that are not sent to ChatGPT in the current context because they are over the token limit.

Delete a chat history

/delete chat <chatname>

Delete all chats

/delete allchats

ChatGPT model

This script defaults to gpt-4. If you do not have access to it or would like to use another version edit this line as needed:

    Model = "gpt-4"

The script will fall back to gpt-3.5-turbo if there is an API error:

        response = None
        try:
            response = openai.ChatCompletion.create(
                model = self.Model,
                messages = messages
            )
        except openai.error.APIError as error:
            # if there is an API error with message exceeded maximum allotted capacity, switch to gpt-3.5-turbo model
            if error.status == 429 and model == 'gpt-4' :
                self.printMessage("GPT 4 is unavailable, switching to GPT 3.5 Turbo", message_from="prompt")
                self.chat(prompt, model = "gpt-3.5-turbo")
            else:
                # if it's another kind of error, raise the error
                raise error

Edit self.chat(prompt, model = "gpt-3.5-turbo") as needed to create a fallback for yourself. Or, you can replace the entire section to to:

    response = None
        response = openai.ChatCompletion.create(
            model = self.Model,
            messages = messages
        )

Change models during chat

You can change GPT models during you chat with:

/model <your preferred model>

History

Chat history is saved in a sqlite file in your home directory, this is required for the chat to work correctly. If history is not saved, then every message restarts the chat and ChatGPT doesn't have any context from previous chat entries.