Tabbit AI Browser: Free, Automated Workflows and Agent-Powered Features

After 100 days of public testing and more than ten version updates, Meituan’s GN06 team released Tabbit 1.0 official version, positioning it as a representative AI‑native browser in China’s AI‑browser market.

Tabbit goes beyond the traditional window model by embedding a large‑model AI, conversational interface, clever tricks ("妙招"), intelligent agents, and scripting capabilities that actively help users understand web content, organize information, and complete actions.

During the public test, users reported that reading papers, which previously required copying, switching tools, and manual comparison for 3–4 hours, can now be finished in about 20 minutes by opening the paper in Tabbit and asking questions on the fly.

The 1.0 release emphasizes upgraded Agent abilities. In the beta phase agents mainly performed simple web operations such as clicking and form filling. The new version expands agents to handle multi‑modal tasks that combine web pages, files, spreadsheets, and PPT templates, enabling information aggregation, document generation, and content delivery.

Tabbit 1.0 standard edition is announced as permanently free, including access to leading domestic large models (MiniMax‑M3, DeepSeek‑V4‑Pro, Kimi‑K2.6). The free quota allows roughly 1,000 conversational turns, 50 image generations, and 10 agent tasks per week, and is available for macOS and Windows.

A professional edition will later be offered at ¥9.9 per week, granting access to more than ten large models for heavy users.

The release also introduces a mobile version for iOS and Android, currently in a trial phase with some features still being refined.

Machine‑Intelligence‑Heart (机器之心) conducted a hands‑on experience, noting that Tabbit’s integration of AI into the browsing workflow makes web reading, information整理, tool invocation, and repetitive actions significantly smoother.

Two types of “妙招” are highlighted: script‑based tricks that generate executable scripts to modify pages (e.g., ad‑blocking, content extraction, style changes) and intelligent‑agent tricks that encapsulate entire task workflows for reuse. The latter can be saved as a “Runtime Skill” and applied to similar tasks such as extracting CVPR 2026 hot topics, where Tabbit automatically searches, reads, aggregates, and summarizes conference data into a structured report.

Compared with traditional browser extensions, which solve fixed needs, AI agents can address ad‑hoc, personalized, and hard‑to‑standardize requirements without the user having to pre‑install specific plugins.

The article places Tabbit within the broader evolution of AI browsers: the first wave began in early 2023 when Microsoft added OpenAI‑powered chat to Bing and Edge; subsequent development split into two paths—enhancing existing browsers (Edge, Opera) and building AI‑native browsers like Tabbit, ChatGPT Atlas, and Perplexity Comet. Recent trends show a shift toward agent‑driven capabilities, turning AI from a passive Q&A tool into an active executor of complex workflows.

Beyond efficiency gains, the author argues that AI browsers help users cope with the overwhelming amount of information on the modern web by automating content summarization, layout cleaning, and personalized presentation, thereby making information more accessible.

Real‑world cases illustrate the impact: a blind piano tuner completed a PPT presentation using Tabbit’s AI‑generated slides; an HR professional streamlined recruitment analysis; a construction engineer reduced a four‑person‑day monthly data‑collection task to two reusable tricks.

Overall, Tabbit exemplifies how AI‑native browsers can empower users to compose their own workflow ecosystems, moving the browser from a static tool to a dynamic platform for personalized automation.