⌨️ 🦾 Zsh Codex

AI in the command line.

You just need to write a comment or variable name and the AI will write the corresponding code.

What is it?

This is a ZSH plugin that enables you to use AI powered code completion in the command line. It now supports both OpenAI's Codex and Google's Generative AI (Gemini). OpenAI Codex is the AI that also powers GitHub Copilot, while Gemini is Google's advanced language model.

How do I install it?

Manual Installation

1. Install the OpenAI package, the Google package, or boto3.

pip3 install openai

or

pip3 install google-generativeai

or

pip3 install boto3

2. Download the ZSH plugin.

git clone https://github.com/tom-doerr/zsh_codex.git ~/.oh-my-zsh/custom/plugins/zsh_codex 

3. Add the following to your .zshrc file.

Using oh-my-zsh:

    plugins=(zsh_codex)
    bindkey '^X' create_completion

Without oh-my-zsh:

    # in your/custom/path you need to have a "plugins" folder and in there you clone the repository as zsh_codex
    export ZSH_CUSTOM="your/custom/path"
    source "$ZSH_CUSTOM/plugins/zsh_codex/zsh_codex.plugin.zsh"
    bindkey '^X' create_completion

4. Create a file called zsh_codex.ini in ~/.config. Example:

; Primary service configuration
; Set 'service' to match one of the defined sections below.
[service]
service = groq_service

; Example configuration for a self-hosted Ollama service.
[my_ollama]
api_type = openai
api_key = dummy_key
model = llama3.1
base_url = http://localhost:11434/v1

; OpenAI service configuration
; Provide the 'api_key' and specify a 'model' if needed.
[openai_service]
api_type = openai
api_key = <openai_apikey>

; Groq service configuration
; Provide the 'api_key'.
[groq_service]
api_type = groq
api_key = <groq_apikey>
model = gemma2-9b-it

; Mistral service configuration
; Provide the 'api_key'.
[mistral_service]
api_type = mistral
api_key = <mistral_apikey>
model = mistral-small-latest

In this configuration file, you can define multiple services with their own configurations. The required and optional parameters of the api_type are specified in services/sevices.py. Choose which service to use in the [service] section.

6. Run zsh, start typing and complete it using ^X!
7. If you use virtual environments you can set ZSH_CODEX_PYTHON to python executable where openai or google-generativeai is installed. e.g. for miniconda you can use:

export ZSH_CODEX_PYTHON="$HOME/miniconda3/bin/python"

Fig Installation

Troubleshooting

Unhandled ZLE widget 'create_completion'

zsh-syntax-highlighting: unhandled ZLE widget 'create_completion'
zsh-syntax-highlighting: (This is sometimes caused by doing `bindkey <keys> create_completion` without creating the 'create_completion' widget with `zle -N` or `zle -C`.)

Add the line

zle -N create_completion

before you call bindkey but after loading the plugin (plugins=(zsh_codex)).

Already exists and is not an empty directory

fatal: destination path '~.oh-my-zsh/custom/plugins'

Try to download the ZSH plugin again.

git clone https://github.com/tom-doerr/zsh_codex.git ~/.oh-my-zsh/custom/plugins/zsh_codex

---

Passing in context

Since the current filesystem is not passed into the ai you will need to either

1. Pass in all context in your descriptive command
2. Use a command to collect the context

In order for option 2 to work you will need to first add export ZSH_CODEX_PREEXECUTE_COMMENT="true" to your .zshrc file to enable the feature.

[!WARNING] This will run your prompt using zsh each time before using it, which could potentially modify your system when you hit ^X.

Once you've done that and restarted your shell you can do things like this:

# git add all files. Also commit the current changeset with a descriptive message based on $(git diff). Then git push

More usage examples

---

Fish Version

Traffic Statistics