ChatGPT CLI

build

ChatGPT CLI is a command-line interface tool for interacting with OpenAI's ChatGPT API. It allows users to communicate with the GPT model, adjust API parameters, use magic commands, insert file contents, fetch content from URLs, and execute system commands while chatting.

Features

Installation

  1. Install Crystal

  2. Build the project:

    git clone https://github.com/kojix2/chatgpt-cli
cd chatgpt-cli
make
sudo make install

  3. Set your OpenAI API key as an environment variable:

    export OPENAI_API_KEY="your_openai_api_key"

Usage

To start using ChatGPT CLI, run the chatgpt command in your terminal:

chatgpt

Type a message and press Enter to get a response from GPT-3.5 Turbo. To exit, type exit, quit, or press Ctrl + D.

Options

You can set various options when running the chatgpt command:

Usage: bin/chatgpt [options]
    -m MODEL, --model MODEL          Model name [gpt-3.5-turbo]
    -i ID, --id ID                   ID of the custom system message
    -s STR, --system STR             System message
    -n INT                           Number of edits to generate [1]
    -t Float, --temperature Float    Sampling temperature to use [1.0]
    -p Float, --top_p Float          Probability threshold of nucleus sampling [1.0]
    -v, --version                    Print version info and exit
    -h, --help                       Print help

Select a pre-defined system command

chatgpt -i code

To add or remove system commands, see the CONFIGURATION section.

File Insertion with %{}

You can insert the contents of a file into the chat by enclosing the file path in %{}. For example, to insert the contents of a file called example.py located in the src directory, use %{src/example.py}.

> Please explain what the following code does. Are there any bugs? %{src/example.py}
> Please read the code of the tool: %{src/**/*} Then update README.md %{README.md}
...
> %w README.md

Web Page Insertion with %%{} (experimental)

You can insert the contents of a web page.

> Please choose five interesting topics from %%{https://news.ycombinator.com/}

Here the HTML from the URL is fetched, the words used in the body are extracted and passed to ChatGPT. It is much smaller than the raw HTML, but still not enough.

Magic Commands

Within the CLI, you can use magic commands to perform various actions:

Note that for %config, %data, and other commands that launch an editor, the editor used can be set by the EDITOR environment variable. The default editor is vim.

The tool is under development, and the magic commands are still being improved.

Executing System Commands

You can execute system commands while chatting by prefixing the command with the ! symbol. For example, to check the current working directory, type !pwd and press Enter. Similarly, you can execute other system commands like !ls, !date, etc. If you want to execute a command and record its output for later use, you can prefix the command with !! instead, then use %STDOUT or %STDERR to insert the captured output into the chat.

Examples:

  1. Execute a command, and display the output immediately:
> !pwd
> !vim
> !htop
  1. Capture the output of a command for later use:
> !! git diff
... git diff output ...
> Please write commit message: %STDOUT
> !!wrong_command
> Explain the meaning of this error message: %STDERR

Execute code blocks captured from markdown in the response

When ChatGPT returns code blocks enclosed with triple backticks, ChatGPT CLI saves these blocks temporarily and assigns them to environment variables named CODE1, CODE2, ... and so on in sequential order. This allows you to execute the code blocks by referencing the environment variables in a system command. For example, if the response contains Python, Ruby, and Bash code blocks, you can execute them using $CODE1, $CODE2, and $CODE3, respectively.

> Write code to display 1 to 10 in Python and Ruby.
> Python:

```
for i in range(1, 11):
    print(i)
```

Ruby:

```
(1..10).each { |i| puts i }
```

> ! python $CODE1
> ! ruby $CODE2

Configuration

The system messages used by ChatGPT CLI can be customized through the config.json file. The file is located in ~/.config/chatgpt-cli/ by default, but it can be changed by setting the CHATGPT_CLI_CONFIG environment variable.

The config.json file has the following structure:

{
  "system_messages": {
    "tran": {
      "role": "system",
      "content": "I want you to act as a translator, spelling corrector, and improver."
    },
    "code": {
      "role": "system",
      "content": "I want you to act as a programmer, writing code."
    },
    "poet": {
      "role": "system",
      "content": "I want you to act as a poet, writing poetry."
    }
  }
}

Uninstallation

rm /usr/local/bin/chatgpt   # Remove the executable
rm -r ~/.config/chatgpt-cli # Remove the config directory
rm ~/.chatgpt_history       # Remove command history

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/kojix2/chatgpt-cli.

Development

git clone https://github.com/kojix2/chatgpt-cli
cd chatgpt-cli
shards install
shards build --release
bin/chatgpt

License

This project is open-source and available under the terms of the MIT License.