Meituan GN06 Launches Tabbit AI-Native Browser

Meituan's experimental GN06 team launched Tabbit, an AI-native browser entering public beta in March 2026 for macOS and Windows. Tabbit reframes the browser as an active assistant, offering conversational context on pages, a background agent mode that can run parallel tasks across tabs, and a Skills system for frequently repeated actions. The product bundles Meituan's LongCat-Flash-Chat while allowing users to switch to GPT, Claude, DeepSeek, Doubao, Qwen, and Kimi models.

Tabbit's core differentiator is its agent-first architecture rather than retrofitting AI into a traditional browser UI. Key capabilities include:

• •Conversational page understanding: the browser parses selected text or an opened article and answers context-aware questions.
• •agent mode: autonomous multi-step workflows that open pages, extract data, fill forms, and synthesize results while running in the background.
• •Skills system: user-defined, no-code macros that convert repeated prompts into one-click commands.
• •Tab and workspace management: automatic grouping of related pages into a vertical sidebar to reduce clutter.

The product supports multiple LLM backends and runtime switching, which lets developers and power users pick models for latency, cost, or capability trade-offs. GN06 has publicly committed to not harvesting user data or serving ads, although implementation details for model telemetry and local vs remote inference were not disclosed. There are also early reports of a code-translation dispute the team addressed by removing a controversial translation, a reminder of integration and IP risks.

Tabbit is a concrete step in the broader agentification trend that shifts value from static UIs to autonomous, workflow-aware assistants. Unlike browser features that surface AI via side panes or search enhancements, Tabbit embeds agent mode as a first-class capability. For practitioners this matters because:

• •It exposes new integration points for LLMs inside multi-domain workflows and cross-site automation, increasing the attack surface for privacy and security issues.
• •Support for multiple models highlights a pragmatic approach to capability heterogeneity rather than lock-in to a single provider.
• •For Meituan, Tabbit could become a strategic on-ramp to local services and instant retail, turning browsing sessions into commercial funnels if tightly integrated.

However, shifting user habits away from passive browsing to agent-managed workflows is nontrivial and will determine the product's reach beyond early adopters.

Adoption metrics, the details of how model requests and user data are routed and logged, how Tabbit sandboxes agent actions for security, and whether Meituan opens APIs or extensions for ecosystem partners. If Tabbit demonstrates reliable, safe automation and compelling integrations with local services, it could change how browsers are evaluated by both users and enterprises.