Personal AI Development Environment

A new approach to building personalized AI-powered development environments focuses on 'live coding' and 'vibecoding', allowing developers to evolve their tools in real-time.

If you're the kind of person that likes tweaking your dotfiles, your tmux config, your vim setup, your VS Code plugins--this is maybe the most fun you'll ever have building something. It's the first time I've felt a sustained flow state while coding with AI.

The goal is to create a personalized AI-powered development environment. Something that works the way you want it to.

The setup described here encourages live coding: you modify the interface while you're using it. This can get a little funky--you may need to drop back down to the command line if you break things too badly. But the live feedback loop is addictive.

This project is released as Open Prompt, rather than Open Source. There are a few reasons:

You'll have more fun this way
My code is hyper-tailored to my own needs
and it's vibecoded mess that happens to mostly work

You don't want my vibecoded mess. You want your vibecoded mess, whose intricacies and quirks and powers only you know.

You can start by giving this README to the coding agent of your choice. Just paste the URL and hit go I guess.

It'll probably do something like:

Read AGENTS.md
Read architecture.md
Send out subagents for each of the subpages
Make you a ~~fully~~~~semi~~probably not yet functional but very ambitious AI-powered development environment

You can wiggum it from there.

What's really fun is that once the app is semi-functional, you can live-code it while you're using it. You can just say "reskin to 1990s hacker style" and watch it morph in real time.

I've tried to architect things to make this easier: all the pages run as separate apps, so you can muck around with one part without breaking everything.

If you're a human you should also read MANUAL.md and learn about a few things you'll have to do by hand. And security. Please think of the security. In case you think security doesn't really matter, here's me catching the agent backdooring its way into the app so it could take the screenshots below:

good luck have fun

Chat UI for creating and streaming agent conversations using the OpenHands SDK.

Kanban board of conversations organized by project and execution status.

Draggable card dashboard for monitoring live conversations in real time.

Settings page for choosing an LLM provider, model, and API key.

Key-value secret store injected into new conversations.

Git repository overview for all projects under ~/git/

Cron-based scheduler that launches agent conversations on a timer.

CRUD editor for agent skills stored in the agentskills.io format.

Configuration editor for Model Context Protocol servers.

Twilio webhook that creates agent conversations from incoming text messages.

Live system monitor showing CPU, memory, disk, processes, and network.