Claude Code is fantastic, it does the same job I wanted to achieve with this. It’s a CLI instead of an UI, but it does 10x or 100x more than the current application situation.
I know, AI is a big, overrated hype on a big percentage of the situations, I’m with you on that one, should that be the case! Fun fact: I write all of these words, I don’t have an AI writing for me!
BUT. A big BUT - pun unintended! There’s no denying that a developer (and other skills) who doesn’t adopt AI will eventually stay behind from one who does so. I’m not saying use an AI to develop an app just by speaking and waiting for the Holy Grail code that will solve all of that. No, no.
Mileage may vary. I think it’s a great fit for repetitive tasks, for errors, especially while learning a new skill, the AI which crunched billions of data points will provide you want you need to know, instead of that being your job fiddling with dozens of websites, forum posts, etc - instantly!
All this to say I’m launching this App, to adhere to what has been said.
TLDR: It’s a Laravel App, that runs natively on your computer (thanks NativePHP), so you just open it, start a new Project a group of chats on a defined project folder - it can be in any language, AI will know!
As of today, you have to clone the repo, and run with artisan serve or Herd , and you don’t have a Projects group, you just give the Project path in an environment variable.
It’s Open Source, free, and you bring your own API key - Anthropic only - aiming for Sonnet 3.5!
Why building this?
There are many tools trying to help in this task, but they’re not helping e in the way I want. So, I built (started building) my own! Nuno on his stream inspired me to open-source it right away - I’d probably take months or if at all!
Examples of what I‘m saying:
I want something hackable with code. I wan to be able to write my logic for stuff. Reuse that. I don’t mean changing the source code.
Even after a Native App (install just like Chrome or Excel for example), I want a place to write some PHP/JS, and change how the scope is given to the AI.
I want to ask something about these 4 files: I have copy-paste them 1 by 1 to another app, or at best, pick them in a tree, mention them all
No, I want to have something like “scopes”:
Scope1 : resources/views/** , resources/js/**
Scope2: app/Models/**database/migrations/**
I can make scopes based on file picking, or defining these patterns/glob/regex.
I can then mention a @scope1 andI want auto magically the thing to be there.
No magic, just a replacement - That I can define how it happens!!!
I want to edit the system instruction, according to the scope I’m using
Hey, AI, you can run php artisan commands , and the list of commands you can run are these, they do such and such, and the arguments are this and that.
When I ask something to the AI, it can now decide to use such Tool and stop it’s message (with a stop_reason = tool_use), so that our software can run that tool for it, and provide the result. Only then the message will be completed later on.
AI, you can run artisan commands - (and provide the output of php artisan list)
E.g.: Like the result of commands: php artisan model:show
AI, you can run bash commands
I may want to send the file tree of the project (excluding vendor, node_modules, .git, etc), and instruct the AI: these are the project files, stop and ask me for any of those if relevant to figure the solution.
The list can go on, and it will! RAG, Model Context Protocol, Agents. One can dream! Talk about scope creep, right??
I just want to be with the feet in the ground, and I’m already making the task of asking the AI, the way I do, in a much convenient way, and with much richer information!
So, like me, many others might benefit from the same solution!
How does it look, what does it do?
We set an environment variable with the project path.
And edit rules files here, add as many as you want.
And from there, we can start!! Just go to localhost:8000 (accordingly to Artisan server or Herd)
While still in development, already usable!
I press # and chose the rule file I want. On the new message, I press @ to refer to the file.
Typically, we just press send. But I want to monitor the things, so I have a mid-step, prepare prompts.
Nothing comes to my mind better than showing on a video!
