About it

This project uses llamafile as a base (Thank you Mozilla!)

This project acts like a middleware to interact with llamafile service. So you can create custom commands that will be used instead of getting the AI results.

Ex: You can create a custom command to check the weather that will call and API instead of getting a response from the AI. You can also create custom commands to interact with databases, etc.

Requirement

• PHP 8.1 or newer
• llamafile service running

How to start

Model

Download a model and save it to the llamafile directory. The model should be in the format of llamafile. You can find models on llamafile repo.

Create a config file

cp config.example.jsonc config.jsonc

Put the downloaded model name in the llamafile field of the config file. Ex:

{
	"llamafile": "llamafile/DeepSeek-R1-Distill-Llama-8B-Q4_K_M.llamafile",
	...
}

If you need any extra system messages, just add it to the system_messages array in the config file.

Check if the other configs are matching with your environment too.

How to create custom commands:

• Create a class inside commands path that implements the interface interfaces\Command
• Implement the execute method
• Use attributes\Description attribute to describe the command. It is REALLY IMPORTANT to be underestood as a custom command, the more clear the description and the class name, the better.

Example:

<?php

namespace commands;

use attributes\Description;
use interfaces\Command;

#[Description('Returns the system username')]
class Whoami implements Command
{
    public function execute(): string
    {
        return shell_exec('whoami');
    }
}

Usage

wlopes@WLopes:~/Project$ php main.php

  - Enter your command here:
Hey!
>>> Hello! How can I assist you today?

  - Enter your command here:
Can you please return my system username?
>>> wlopes

  - Enter your command here:
how to print a 'hello world' in python?
>>> To print 'hello world' in Python, you can use the print function. Here's how:

print('hello world')

In the snippet above we can se a normal AI response and also a custom command returning the system username.

TODO

• Add more default commands
• Add a custom path that will be git ignored for user commands
• Change how it is handling llamafile to allow it to connect to other APIs too
• Voice commands one day? 👀