Meituan's GN06 Team Launches AI Browser Tabbit 1.0
N.R. Finch
Meituan's GN06 team has launched Tabbit 1.0, an AI browser whose core pitch is letting AI read everything in your browser directly — no more copy-pasting between tools. The pro tier costs ¥9.9/week; basic features are free forever.
Why build AI into a browser?
Tabbit's core logic: a user's tabs, web pages, and local files already live inside the browser — for AI to do real work, it needs to see what you see.
This means → embedding AI in the browser eliminates the copy-switch-paste-switch-back loop between a standalone AI tool and your workspace.
In plain terms = you used to ferry information to AI; now AI reads the screen itself.
What can it actually do?
Context direct-read: summon an AI chat on any web page; the AI answers based on the page content. Users can highlight text or screenshot images as references, and use @ to pull in content from multiple open tabs at once.
Local file handling: @ a local folder and Tabbit switches to Agent mode — an AI-driven task-execution mode — to process files directly, no upload required.
Multi-model parallel runs: Tabbit connects to major Chinese LLMs. Users can select several models to answer the same question side by side, with built-in comparison.
What is the "Tricks" feature?
"Tricks" come in two types: prompt tricks (preset AI instruction templates) and script tricks (AI writes code on the spot to modify the web page you're viewing).
Script trick examples: disguise a page as a spreadsheet, auto-generate a table of contents for an article, or batch-export comment data — each trick can be reused with one click.
The platform also offers a "Tricks Plaza." In plain terms = skills other users have built are ready for you to grab and run — no coding needed.
How does it handle Agent tasks?
Tabbit automatically isolates Agent runtime into a separate tab group, keeping it apart from the pages you're browsing.
This means → unlike some desktop agents that take over the full screen, Tabbit lets you keep working while AI runs tasks in the background.
Where does the product stand — and who else is in this race?
Tabbit has been in public beta for 100 days across 12 iterative releases. Desktop clients ship on Windows and macOS; mobile betas are live on Android, iOS, and HarmonyOS app stores.
Competition is heating up: OpenAI is folding its AI browser Atlas into a "super-app" plan, and Perplexity has already shipped Comet.
This reflects a shared bet that the browser is the critical gateway for AI — but user browser habits are deeply entrenched. Whether Tabbit can build retention is the key test of its commercial viability.
Content is for reference only, not financial advice.