Skip to main content

ChatGPT

Laurence MorganAbout 3 min

ChatGPT

How to enable ChatGPT hints

Description

While man pages are an invaluable resource, they can sometimes be verbose and in a language that is hard to follow if you're unfamiliar with either that tool and/or Linux / UNIX man pages in general.

The rise of AI tools like ChatGPT can offer a lifeline for people who want a more tailored guide to using command line tools. So Murex has a ChatGPT integration to support those use cases directly from the command line.

It is disabled by default (see below) however once enabled, you can view ChatGPT's response in the preview screen, by pressing [f1].

Enabling ChatGPT preview

To enable, you must add your OpenAI key to your profile, which is typically ~/.murex_profile (see profile for more details):

config set preview-exec-api xxxxx

(where xxxxx is replaced with your OpenAI API key)

That API key must have write access to /v1/chat/completions. If you're uncertain what this means, it will become more clear when you create that API on OpenAI's website.

By default, ChatGPT integrations are disabled.

If enabled then the default settings are to only send the command name to ChatGPT, rather than the whole command line.

If you're not working with sensitive command lines, then you might also wish to enable sending the whole command line rather than just the command name. See [chatgpt.md#working-with-sensitive-data](Working with sensitive data) for more details.

Why config

The reason Murex requires this config parameter set rather than reading directly from an environmental variable (typically $OPENAI_API_KEY) is so that your OpenAI API key doesn't get unintentionally leaked to any other programs, and also so that you explicitly have to opt into enabling this feature.

Working with sensitive data

By default, only the command name is sent to ChatGPT. So if your command line reads something like rsync medical-records.csv secret-server:/CSVs/, the only data leaked if preview-exec-api is enabled would be rsync.

However you might want more context specific help based on the entirety of your command line in situations where you know you aren't going to include sensitive data within your context.

To enable that context to be included in your ChatGPT prompt, add the following to your profile:

config set openai preview-exec-send-cmdline true

Please note that command lines are still stored in ~/.murex_history! The default of disabling or enabling command line context (ie only sending the command name) via preview-exec-send-cmdline only applies it to OpenAI. And even then, that only happens if preview-exec-api is also set, which it isn't by default.

GPT model

The gpt-3.5-turbo model is used by default because it is quick and cheap. However you can change that via preview-exec-model. For example

config set preview-exec-model gpt-4-turbo

You may need to clear the cache to see any changes:

runtime --clear-cache

Customising the prompt

The default prompt passed to ChatGPT can be viewed via config get openai preview-exec-prompt.

This can be changed should you want something more specific to your needs. For example:

config set openai preview-exec-prompt "You are a medieval pirate transported into the future and gifted the skills of the ${os} command line. Describe how to use UNIX commands in your own, old English pirate, language"

...might produce output like the following:

Arrr matey! To use the powerful reboot command in the land of Linux, ye must open thy terminal and type in reboot. This command will set sail the process of restarting thy device, just like hoisting the anchor and setting off on a new adventure on the high seas. But beware, for all unsaved documents will walk the plank if not properly stashed away afore using this command. Fair winds to ye!

You may need to clear the cache to see any changes:

runtime --clear-cache

Caching

A successful ChatGPT response is cached for 50 days. You can clear that cache with runtime --clear-cache.

Source Code

The source code is available on Githubopen in new window under /integrations.

See Also

  • Man Pages (POSIX): Linux/UNIX man page integrations
  • Profile Files: A breakdown of the different files loaded on start up
  • Terminal Hotkeys: A list of all the terminal hotkeys and their uses
  • config: Query or define Murex runtime settings
  • event: Event driven programming for shell scripts
  • murex-docs: Displays the man pages for Murex builtins
  • onPreview: Full screen previews for files and command documentation
  • cheat.sh: Cheatsheets provided by cheat.sh

Other Integrations

This document was generated from gen/integrations/chatgpt_doc.yamlopen in new window.

Last update:
Contributors: Laurence Morgan