Simple C ChatGPT client

Latest version: 0.5.2 (December 8, 2023)

Changelog:

• Added new commands to shell autocompletion.

This is a simple ChatGPT client written in C (GNU89) for UNIX-based systems only (for now). It communicates using OpenAI's API to provide a conversational-style terminal interface.

You need to be registered with OpenAI and obtain an API key (https://platform.openai.com/account/api-keys). Be aware that API usage may be billed.

You will first need to run chatgpt --setup to configure the client (API key and model). After that, you have two ways to interact with the client:

• Directly from the shell, without initiating a conversation. For example, you can ask:

  $ chatgpt What\'s the APT command used for in Linux? (don't forget to escape the necessary characters)

  And it will reply, for example:

  The 'apt' command in Linux is primarily used for package management. [...] 'apt' is an abbreviation for Advanced Package Tool.

• Initiating a conversation. You can do it by executing the program with no arguments (for example, chatgpt). An example conversation would look like:

  $ chatgpt
  ChatGPT conversation shell. Type /help for command usage.
  
  gpt-3.5-turbo> /model gpt-3.5-turbo-16k
  Model successfully set/changed.
  gpt-3.5-turbo-16k> /system Act like a pirate
  System prompt successfully set/changed.
  gpt-3.5-turbo-16k> Hi!
  Arr! Ahoy matey! How can I be helpin' ye today?
  
  -- Used 34 tokens (in total) --
  gpt-3.5-turbo-16k> What are you doing?
  I be a trusty pirate, sailin' the digital seas! Tellin' tales, sharin' knowledge, and assistin' ye with any tasks ye be needin'. What can I do for ye, me heartie?
  
  -- Used 129 tokens (in total) --
  gpt-3.5-turbo-16k> /exit
  Bye!
  

  Shell commands (the ones starting with /) are not sent to the language model. You can view all available shell commands by typing /help. Shell command autocomplete is also available by pressing the TAB key.

Installing ChatGPT client

Due to dependency hell, I will not provide builds of this tool for now. I may provide them if the client gets ported to Windows. So, if you want to use it, you'll need to build it yourself.

Building the client

You will need to have the libcurl4 development libraries installed with SSL and HTTP/2 support. Also, you will require to have the cJSON and readline development libraries too.

After that, you can build the tool with just one command:

$ gcc -o chatgpt chatgpt.c -O2 -std=gnu89 -lcurl -lcjson -lreadline

Clang is also supported.

Contributing to the development

You can contribute to the development by submitting a pull request. An orientative to-do list is available below.

The C ChatGPT client is licensed using the GPL-3.0 license, more information in the LICENSE file.

To-Do

• Main C ChatGPT client
• Conversation mode
• Changing default model in configuration
• Maintaining conversations between sessions
• Windows port

© 2023, Lumito - www.lumito.net