Thank you. It will be great if you can post it for me:

Title: Show HN: A Desktop App That Lets AI Control Your Browser For You

link: https://github.com/upwindchange/Autai

Hey everyone,

I've been working on *Autai* — an open-source desktop app (Electron) that uses AI agents to automate your browser. You just type what you want, and the AI opens a real browser and does it for you.

Beyond that, this software can be used as a normal LLM chat interface, similar to cherry-studio.

*Links:*

GitHub: [https://github.com/upwindchange/autai](https://github.com/upwindchange/autai)

Downloads: [https://github.com/upwindchange/autai/releases](https://github.com/upwindchange/autai/releases)

*Why this project:* When I saw the Browser-Use project (the python one), I had a "Wow" moment and tell myself, this would be the future of AI helping people doing everyday task. But when I tried it, it feels very geek to me: you have to use chrome browser via CDP debug mode and the python code runs in a terminal. I was thinking if it is possible to build something that can provide the user an easy to use version of "browser-use" with a nice interface and a nice human in the loop interaction?

I went through a lot of exploration, e.g. build upon cheery studio or lm studio, integrate the python runtime and the whole browser-use project into it. At the very end, every shortcut takes a tradeoff and I decided to write my own from scratch.

So here I am after close to one year of engineering and trials: the first POC release. I wrote the front-end using assistant-ui, back-end using ai sdk. A ton of other packages are used, you can have a look at the package.json file in the project.

Relationship to Browser-Use: You can treat this project as a Typescript rewrite with its own agent, own front-end, etc. More precisely speaking, the CDP tool of browser-use project is rewritten into the Typescript CDP tool in this project. I wrote everything else.

This is THE project in which you won't call an ELECTRON app bloated: because I am using it as a browser, making full usage of its blink + node + CDP debugger engines.

*What it can do:*

- *Browser Automation* — "Add these items to my Target cart", "Book a flight from NYC to SF on Friday", "Fill out this form" — the AI plans the steps and executes them in a real browser session - *Research Mode* — Ask a question and it searches the web, reads multiple sources, and gives you a synthesized answer. No more 20-tab skimming - *Multi-session* — Run multiple browser tasks in parallel - *100+ AI providers, 4,000+ models* — Works with OpenAI, Anthropic, Google, DeepSeek, xAI, Ollama (local), and many more. Bring your own API key

*You stay in control:*

The AI pauses for CAPTCHAs, logins, and payments and hands control back to you. There's a split-view so you can watch everything the AI does in real time.

*Other nice touches:*

- Auto-tagged conversations with search and filtering - Syntax highlighting, math rendering, Mermaid diagrams in AI responses - Image and file attachments - Dark/light mode - English and Chinese UI

*Project status:*

Autai is in *active alpha development* and evolving fast. I'm heads-down building right now, so issues and feature requests are closed for the alpha phase — they'll open up once it reaches beta. That said, feedback and thoughts are always welcome here in the comments.

MIT licensed. Happy to answer q...