Since vibe coding became a thing I’ve been more productive than ever, shipping multiple side projects in weeks or even days (such as https://news.ycombinator.com/item?id=46196796 :D). But as I built more complex apps, I feel like there's something missing.

Cursor is powerful, but it felt cumbersome and heavy for my workflow — bloated with features I didn't need. I actually really like Lovable for its simplicity, but I also want to have more control of my code and architecture decision. I wanted something in the middle: a tool for developers that offers a seamless AI agent experience directly within Next.js/React, without the constant window switching.

So, I decided to build it myself. I started it with https://github.com/MarsWang42/Awel

I want a model-agnostic AI coding agent designed to keep myself in the flow. It acts as a bridge between models (most importantly, Claude Code) and your browser. It understands what is happening in the browser context without relying on flaky MCP or skill solutions. And most importantly, let me interacts with web elements, make screenshots and annotates, then send to the agents.

It's really simple to use. If you have Claude Code already setup, just run `npx awel dev` in your Next.js projects, or you can setup other models by providing api keys in your env variables.

What feature do you think I should implement next? I’m building this in public and want this to be a community effort.I’d love for you to try it out, break it, and let me know what you think. Hope this workflow resonates with you