Like Clippy but for the CLI. A blazing fast AI helper for your command line.
Clipea is a streamlined, cheap and hackable tool that integrates GPT with your console. It's like Github's Copilot for CLI but it's cheaper, faster and doesn't get in your way.
Tell Clipea what you want to do and it'll give you a shell command, asking you if you want to run it. Clipea works even better with Zsh, as it adds the shell command to your console as a pending command, just as if you had typed it yourself!
Clipea was born out of Hubcap, an experimental autonomous agent. Clipea is less dangerous and actually designed to be a usable productivity tool, rather than just a tech demo.
Advantages over copying and pasting from ChatGPT:
- Speed. No need to switch to your browser and back
- Shell integration, like filename completion
- Automatically knows things like your shell and OS
Warning
AI isn't perfect. Clipea might suggest a dangerous command. Be careful.
brew tap dave1010/clipea
brew install clipea
clipea setup
clipea alias
Once installed, just ask Clipea for a command:
clipea convert test.mp4 to a gif
Type y<enter> to run the command. Anything else or <ctrl-c> to cancel.
Or, if you have the Zsh integration (highly recommended), it's even easier:
?? how many gig free do i have
Just press <enter> to run the command or <backspace> if you want to edit it.
The Zsh integration is more than just a quick alias. It allows Clipea to put the command in your input buffer, ready to run, just as if you'd typed it yourself.
# setup
clipea alias
# usage
?? how many gig free do i have
Benefits:
- Quicker to type
- Shows up in your shell history
- Allows editing with normal shell commands like
<ctrl-c>to cancel or<ctrl-a>to go to the start of the line - Runs the command as a child process of your shell, rather than a child of Clipea
Behind the scenes this is using zsh's print -z.
Just start the query with a "4". Remember that OpenAI charge lots more for GPT-4. Generally the standard GPT-3.5 is fine for commandline stuff.
?? 4 create a text file explaining quantum mechanics in a haiku in the style of a pirate
You can also send in data via stdin. Clipea limits you to 8192 bytes, so the LLM isn't overwhelmed.
ls -F | ?? explain this project setup
Most of the time it's best to give Clipea a filename to create a command for, rather than the actual file contents.
cat package.json | ?? how many packages here # GPT isn't great at counting
?? count how many packages are in package.json # GPT is good at suggesting commands
Clipea gets given some environment limited information like your OS, shell and path. This allows it to give better responses.
?? wheres my shell config
?? install curl
?? compare README.md to my clipboard
You can customize your prompt by creating a prompt file in your home directory.
~/.config/clipea/system-prompt.txt
You might wanna start by copying the default prompt:
cp $(dirname $(readlink -f $(which clipea)))/system-prompt.txt ~/.config/clipea/system-prompt.txt
Then edit ~/.config/clipea/system-prompt.txt to your liking, but be aware that environment info is always added to the end of the prompt.
Just use your shell history by pressing the <up> arrow key. Your cursor will be at the end of
the last line, ready to edit it.
For example: typing ?? list js files recursively may give
$ find . -name "*.js"
Then to edit, press <up> then ignore node modules to get something like
$ find . -name "*.js" -not -path "./node_modules/*"
Clipea doesn't have any context of what it said before, though this may be added in the future if there's use cases it helps with.
?? open my shell login script in my editor
?? Extract package.tar.gz
?? Install something that converts pdf to text
?? Make a 30 char password
?? Find files bigger than 10mb
?? Rename all txt files space to underscore
?? Convert file.avi to gif
?? Decrypt data.txt.gpg
?? Highlight URLs in index.html
?? Show me just the headings from README.md
?? count loc recursively
?? Find replace all PHP files in project that call eval function with safe_eval
?? git fetch, rebase master, safely force push
?? turn orders.csv into sqlite
?? count payments in orders.db
?? open bbc news
?? check the spf record for example.com
?? What port is my webserver listening on
?? Check cors headers for api.example.com
?? Where is nginx writing logs
?? Quick http server
brew tap dave1010/clipea
brew install clipea
clipea setup
pip install llm
git clone https://github.com/dave1010/clipea.git
cd clipea
./clipea setup
./clipea add current dir to my path on shell login
[!INFO] The
??shell alias is highly recommended if you use zsh
clipea alias
Clipea is currently written in PHP but may switch to Python (#3).
Clipea uses llm to interact with large language models.
By default it will use OpenAI's GPT-3.5 model but can be configured to other models, such as Llama.
Running clipea setup will talk you through getting OpenAI keys.
Always read and check what Clipea suggests before accepting it.
Clipea uses OpenAI's APIs by default, though can be set to use any LLM that llm supports.
Only very basic environment info like your OS and editor is sent to the LLM.
Run clipea env to see the data the LLM gets.
As a very rough example, using the default GPT-3.5, 100 Clipea queries to OpenAI cost $0.02. Set a quota and keep an eye on costs to make sure.
Contributions welcome.
MIT License
Copyright (c) 2023 Dave Hulbert