ChatGPT-TUI

A terminal user interface for ChatGPT with an empasis on importing/exporting code (no more copy/pasting!).

Should work for Node.js 14.15.0 and above.

Features:

• 🖨️ Export code blocks to files or copy to clipboard
• 💾 Load code from files and insert into the conversation
• 🌎 Load content from websites (i.e. documentation) and insert into the conversation
• ✒️ Syntax highlighting for code blocks

Why not just use GitHub Copilot?

You should! But one missing feature is the ability to iterate on a solution. For non-trivial problems, its unlikely that the first solution it comes up with is correct. Having the model "think step by step" can dramatically improve its abilities on complex tasks. With ChatGPT-TUI, you can iterate on your solution and export it to a file (without all the copy/pasting).

Quick Start

To install and run, simply type the following into your terminal:

npx chatgpt-tui

If this is your first time running the program, it will prompt you for your OpenAI API key. After you've created an OpenAI account, you can get one here.

From there, you can start chatting with ChatGPT, much like you would with the ChatGPT website.

Loading Local Files

You can load local files into the conversation by using the $FILE(<path>) syntax. For example, if you wanted to load the contents of ./src/index.ts into the conversation, you would do:

The contents of ./src/index.ts are: \n\n $FILE(./src/index.ts)

To load specific lines of a file (as opposed to the entire thing), add brackets after the path but before the closing parenthesis. For example, to load lines 1-5 of ./src/index.ts, you would do:

This will load the first 5 lines of ./src/index.ts \n\n $FILE(./src/index.ts[1:5])

To load an entire directory, you can use the $FOLDER(<path>) syntax. Note, this is not reccomended because you will likely exceed the token limit.

Loading Content from Websites (experimental!)

You can load content from websites into the conversation by using the $URL(<url>) syntax. For example, if you wanted to load the contents of the OpenAI documentation into the conversation, you would do:

This feature is still a work in progress, bugs are expected.

CLI Options

--user-msg

Preload the first message you type. When combined with some bash scripting, this can be used to automate the refactoring of multiple files, especially when the refactoring is non-trivial (i.e. hard to solve with a simple codegen)

For instance, to convert all the files in ./src to use ES6 modules, you could do:

for file in ./src/\*; do
  npx chatgpt-tui --user-msg "Refactor this code to use ES6 modules $file \n\n \$FILE($file)"
done

Notice we had to escape the second $ character in the user message. This is because bash will try to perform variable substitution, which we don't want, because we want it to be interpreted by ChatGPT-TUI.

--model

By default, we use the "gpt-3.5-turbo-0301" model. If you have access to the GPT4 models, you will greatly benefit from using those models.

--debug

Prints out the user message post transformations (i.e. after $FILE substitutions are made).

Future Features

I hacked this together in a few days, so it's not very polished. Here are some features I plan to add:

Support exporting multiple code blocks

Currently, the user has to select each code block individually to write to a file. I'd like to add a feature to export all code blocks at once.

Add CLI options

--log-dir

Should log the entire conversation to a directory.