Turn Your Browser into an AI Assistant: How Tabbit Multiplies Daily Work Efficiency

「Core question this article aims to answer: With all the time we spend in browsers processing information, can AI genuinely help beyond being just another chat window that requires constant copy-pasting?」

For the past six months, I’ve been searching for an answer to how AI should truly integrate into daily work. Recently, I started using a browser called Tabbit, and it made me realize a fundamental issue: we’ve been trying to force AI to adapt to our tools, rather than designing tools that adapt to AI.

Tabbit, developed by the Light Years Beyond team, is arguably China’s first natively built AI browser, redesigned from the ground up. After using it for a while, I’ve found it solves a core pain point: the browser itself is where we consume most information, yet previous AI tools required manually copying and pasting content over to them.

What is Tabbit: A Browser That Actually “Understands” What You’re Viewing

「Core question this section aims to answer: What fundamentally sets Tabbit apart from other browsers or plug-ins?」

Simply put, Tabbit isn’t just a Chrome clone with an AI skin, nor is it just a browser with some extensions installed. It’s designed from the ground up with AI as an integral part of the browser itself.

Five core functions form its foundation:

  1. 「Chat」 – Ask questions while browsing, supporting @webpage, @tab groups, and @favorites
  2. 「Skills」 – Turn repetitive tasks into one-click operations, supporting variables and scripts
  3. 「Agent」 – AI automatically completes complex tasks with full visibility
  4. 「Favorites」 – Semantic search for automatically organized, fragmented collections
  5. 「Tab Groups」 – Automatic tab grouping with cloud synchronization

Each of these features might have been attempted separately before, but their combination within a browser creates a completely different effect. Crucially, it’s deeply adapted for Chinese websites. WeChat Official Accounts and Bilibili function normally, and data can be stored locally.

Case Study 1: From Information Overload to Polished Draft, Time Cut from Half an Hour to Three Minutes

「Core question this section aims to answer: What’s the fastest way to turn a dozen open tabs into a coherent report?」

Recently, the news about Anthropic and the U.S. Department of Defense sparked widespread discussion. I opened over a dozen related tweets and official statement pages, mixed in with other unrelated content. Before, I’d have to read through everything, copy links, open an AI tool, paste them, and ask the AI to read them – the problem being many AIs can’t even read tweet links.

Tabbit’s approach is completely different.

Click the tab management button in the top-left corner, and it organizes all your tabs with one click. It doesn’t just group them by domain name; it genuinely groups them based on content. All Anthropic-related tweets and official documents were grouped together, while unrelated tweets remained separate.

With the groups organized, I went to the AI chat interface on the new tab page, directly @-selected this group, and asked it to compile the event timeline and different stakeholder perspectives. It first explained the background of the controversy, then outlined the stances of the main players, followed by reactions from other players in the industry – each point with a source.

「Reflection:」 Previously, the most time-consuming part of this kind of information整理 wasn’t writing the report, but organizing the sources. Which link said what? Which viewpoint came from where? Just sorting that out took half an hour. Now it’s three steps: one-click grouping → one-sentence query → three-minute draft. This isn’t about replacing thinking; it’s about minimizing the grunt work so I can get to the actual thinking part faster.

Following this logic, you can extend the use case:

Scenario Action Output
Student writing thesis @tab group to organize core ideas Literature review outline
Competitor analysis @tab group to compare features/pricing Comparison report
Travel planning @tab group to summarize attractions/food Itinerary

A general prompt template could be: 「@tab group Organize the [core content] from these webpages, categorize by [dimensions], and generate a [format].」

Case Study 2: Hands-on with the Agent – From Booking Flights to Finding Hotels Automatically

「Core question this section aims to answer: Can we delegate those tedious but logically simple web operations to AI?」

I’m a bit anxious about flying and have specific requirements for aircraft models. Combined with time and location constraints, finding suitable flights is a hassle. On sites like Ctrip, just locating the filter buttons and setting conditions takes a few minutes.

Tabbit’s Agent mode can handle this. After enabling “Agent” in the chat sidebar or on a new tab page, the controlled webpage shows a colored gradient overlay. You can see the mouse cursor movement and upcoming actions, and you can continue browsing other tabs without interrupting its work.

I input a voice request: “Find me flights next month from Beijing to Kunming, not Boeing 737 Max, preferably a large aircraft, departing in the morning and arriving in the afternoon.” After giving the task, I ignored it. When I came back later, it had already completed the筛选 and presented flights (direct and connecting) matching my requirements.

Next, I asked it to find hotels near popular attractions in Kunming. This time, it not only clicked filters but also proactively typed in “Nanping Street” to search.

「Reflection:」 It is admittedly a bit slower than doing it yourself. But the key difference is between consuming time and occupying time. When I do it myself, I’m locked in, staring at the screen, attention fully occupied. Handing it off to an agent frees me to handle other things. For those non-urgent, procedurally fixed but necessary operations, this asynchronous processing fits human workflow much better.

Case Study 3: Turning Operational Workflows into Reusable Skill Templates

「Core question this section aims to answer: Can we save those repeatedly typed instructions as templates for one-click reuse next time?」

The travel planner was handy, but for the next trip to Chengdu or Shanghai, did I have to re-enter the entire prompt each time?

Tabbit’s “Skills” feature solves this. Go to the lightbulb icon in the top right to access Skills and create a new one. Select the Agent category and save the variables from your previous prompt, like time, location, and hotel requirements.

For example, structure the prompt like this:

Help me find flights from {{Departure City}} to {{Destination}} on {{Date}} with aircraft requirements {{Aircraft Type}}. Filter and list the options.

Next time, select this Skill in Chat. It will automatically launch Agent mode, and you only need to fill in the variable values; defaults are used for anything left blank.

You can even design more complex templates:

「Multi-city Price Comparison」:

Help me compare flight prices from {{Departure City}} to {{Destination1}}, {{Destination2}}, {{Destination3}} on {{Date}}. Find the cheapest options. Requirements: 1. Show cheapest flight for each destination 2. Compare total cost (flight + hotel) 3. Generate a comparison table.

「Round-trip Itinerary Planning」:

Help me plan a round trip from {{Departure City}} to {{Destination}}: Departure {{Departure Date}}, Return {{Return Date}}. Aircraft requirement {{Aircraft Type}}, hotel standard {{Hotel Star}} near {{Landmark}}. Requirements: Find optimal flight combinations, recommend hotels for the stay duration, calculate total cost.

A few tips from experience writing prompts:

  • Use Chinese variable names for easier understanding within the context.
  • Set reasonable default values to minimize modifications.
  • If a certain step is consistently inaccurate, adjust the prompt periodically.

Case Study 4: Favorites No Longer Gather Dust – Semantic Search Even for Images

「Core question this section aims to answer: What’s the point of bookmarking thousands of items if you can never find them when needed?」

We all have massive bookmark collections. Most of the time, they’re saved and forgotten. When you need them later, you only remember the gist, not the exact title, making traditional search useless.

Tabbit’s approach: when you favorite a webpage, it reads the content and generates a summary. This allows you to find it later in Chat using natural language, even if you can’t recall the specific title.

What surprised me more is its support for favoriting images, which also get text annotations. For designers, this is a game-changer.

Right-click on an image you like, select “Favorite this image,” and Tabbit automatically analyzes and generates a description of the visual content. In the Favorites section, you see all images with their corresponding text descriptions.

Next time you need an image, just tell Chat: “Find me UI design drafts that are dark-themed and tech-oriented,” and it will locate relevant material.

A more advanced use: @Favorites Find all my dark-themed, tech-oriented UI designs. Analyze their common design language and suggest a color palette.

「Reflection:」 Currently, Chat can’t directly read the image content itself from the Favorites, so the design suggestions generated might deviate from the actual images. I’ve provided this feedback to the team; if they solve this, I might manage my entire素材库 exclusively with it. Even so, just being able to find saved content with natural language already solves the core problem of “favorites = oblivion.”

Case Study 5: Reshape Any Webpage to Your Liking with Scripts

「Core question this section aims to answer: Besides installing extensions, how else can you customize webpage functionality?」

Besides Agent, Tabbit’s Skills have two other categories: Prompts (essentially reusable prompt templates) and Scripts.

The Scripts feature is incredibly powerful – you can modify any webpage according to your needs, adding new functionalities.

When creating a Skill, select “Script,” fill in the name, description, target webpage URL, and the core part: the script code.

For example, common issues when reading articles on WeChat Official Accounts: font too small, tiring to read; ads distracting; no table of contents for long reads. I had Tabbit write a “Reading Enhancer” script that, with one click, provides:

  • 「Visual optimization:」 Eye-friendly colors, large font (18px), comfortable line height (1.8), centered content with max-width 800px
  • 「Functional enhancement:」 Auto-generated table of contents, reading progress bar, back-to-top button
  • 「Distraction removal:」 Removes ads and unrelated elements, hides sidebars and recommendations

The before-and-after对比 is stark; the reading experience is completely transformed.

Another example is enhancing video sites. On Bilibili, playback speed options are limited (0.5x, 1x, 1.25x, 1.5x, 2x). You can’t get 2.5x. There’s no built-in screenshot tool for capturing精彩 moments, and no picture-in-picture mode. I had a script written that adds four buttons to the right side:

  • 「Speed options:」 9档 from 0.5x to 3x
  • 「Picture-in-Picture:」 Video floats over other windows
  • 「Screenshot:」 Capture current frame, preview, download
  • 「Loop:」 Toggle video looping on/off with status indication

Beyond these, possibilities are vast: turn any site into dark mode, one-click download all images, extract emails/phones from pages, auto-fill forms, block ads and pop-ups.

「Reflection:」 Don’t know how to code? No problem. Click “Generate script via Chat” below the input box, describe your requirements in natural language, and AI generates the code. Just paste it back. If the built-in model’s output isn’t great, I’ve created a “Script Writing Prompt Template” you can use with Claude or Gemini to generate the code, then paste it back here.

Why Not Just Use Overseas Alternatives?

「Core question this section aims to answer: With existing AI browsers abroad, why do we need Tabbit?」

Overseas products simply don’t work well with Chinese websites. They can’t read WeChat articles, can’t process Bilibili danmu, and face various access restrictions within China. Tabbit supports switching between multiple models (currently all free), is deeply adapted for Chinese sites, and allows local data storage. For enterprise users, these are non-negotiable requirements.

Practical Action Checklist

If you want to give it a try, follow this sequence:

  1. 「Try Tab Organization First:」 Open a dozen related pages, click the one-click organize button in the top-left, and see how AI understands the content.
  2. 「Try Information Synthesis:」 In Chat, @-mention a tab group and ask it to summarize. Assess the output quality.
  3. 「Experiment with Agent:」 Find a site with fixed procedural steps (booking forms, data entry) and let it handle the operation once.
  4. 「Create Your First Skill:」 Save a frequently used prompt as a template, setting up variables.
  5. 「Play with Favorites:」 Favorite a batch of images. A few days later, try finding them using natural language search.
  6. 「Experiment with Scripts:」 Find a commonly used site with unsatisfactory functionality. Let AI generate an enhancement script for it.

One-Page Summary

  • 「Tabbit is a natively built AI browser,」 not a plug-in wrapper. Core functions: Chat, Skills, Agent, Favorites, Tab Management.
  • 「Automatic Tab Grouping:」 Intelligently organizes based on content, not just domain names.
  • 「Chat can @-mention context:」 @tab groups, @favorites, @webpages, letting AI understand what you’re currently viewing.
  • 「Agent operates webpages:」 Visible and controllable, can handle tasks like bookings and form-filling.
  • 「Skills + Variables:」 Turn repetitive tasks into templates; execute with one click after changing a few variables.
  • 「Favorites with Semantic Search:」 Even images get text descriptions; find saved content with natural language.
  • 「Scripts enhance any webpage:」 Even non-coders can have AI generate code to add features.
  • 「Deeply adapted for China:」 Works with WeChat Official Accounts and Bilibili, local data storage, multiple model options.

Frequently Asked Questions

「Q: Is Tabbit free?」
A: Yes, it’s completely free currently. Multiple built-in models are available to switch between.

「Q: How is it different from a browser with extensions?」
A: Tabbit is designed from the ground up with AI as a core component. It genuinely understands webpage content in a way that simple extensions cannot achieve.

「Q: Is it safe for the Agent to operate webpages?」
A: The entire process is visible. You can see the mouse cursor and upcoming actions, and you can interrupt at any time.

「Q: Can I really search for favorited images?」
A: Yes. Tabbit automatically analyzes images and generates text descriptions, allowing natural language search. However, currently, the Chat feature cannot directly analyze the images themselves.

「Q: Is the Scripts feature difficult to use?」
A: Not at all. You can use natural language to have AI generate the script – no programming knowledge needed. If the built-in model’s output isn’t ideal, you can generate scripts with other AIs and paste them back.

「Q: Is the tab grouping really intelligent?」
A: Yes, it groups based on the actual content of the pages, not just domain names. For instance, different sources about the same event will be grouped together.

「Q: Can I sync favorites from other browsers?」
A: The main focus is on using Tabbit’s built-in favorites feature. It’s recommended to gradually migrate to experience the value of semantic search.

「Q: Is access stable within China?」
A: Tabbit is designed specifically for the Chinese internet environment. It requires no special network setup and has excellent support for domestic websites.

「Q: Where can I download it?」
A: You can search for “Tabbit browser” to find the official download page. Start with a simple task – organize a messy set of tabs and see how it feels. That’s the best way to understand its value.