Ant Group이 GPASS 프레임워크를 AI 안경과 지능형 엣지 단말을 위한 Agent 기반 OS인 Lingying으로 업그레이드했다.

On September 11, at the 2026 Inclusion·Bund Conference "AI-Driven New Terminals and Services" forum, Ant Group announced that GPASS intelligent terminal trusted connection technology framework has officially upgraded to "Lingyinx."

Lingyinx is positioned as an Agent-native operating system and open platform for new terminals such as AI glasses, providing foundational capabilities around natural interaction, intelligent agent execution, multi-device coordination, and trusted security, while opening these capabilities to chip partners, hardware manufacturers, and developers.

Zhou Zhifeng, senior vice president of Ant Group, stated that AI terminals are transforming human-computer relationships, with devices shifting from awaiting user operations to understanding user intent and coordinating services. "Ant Group doesn't make hardware, but we are willing to be the one who 'paves the road.' We hope to consolidate identity, security, payment, and end-cloud collaboration capabilities into infrastructure, allowing manufacturers and developers to focus more on product and scenario innovation."

In the past, the core task of terminals was to present information to users, who actively operated devices through screens, keyboards, and touch controls. The arrival of AI is changing this relationship: devices are beginning to understand what users see, say, and their environment, and are further proactively invoking agents and services to complete tasks.

New terminals such as AI glasses, earbuds, and smart rings are becoming important carriers of this shift. Interaction interfaces are evolving from keyboards and touch screens toward voice, vision, gesture, and eye-tracking, while terminal competition is shifting from capturing user attention to understanding user intent, connecting digital services, and completing real-world tasks. This means AI terminal competition is no longer just about hardware specifications—it also requires new software and service infrastructure.

In 2017, Alipay envisioned through the "Shadow Plan" a form of service freed from mobile phone constraints: a user takes a glance, speaks a sentence, and the service naturally happens. In recent years, as AI terminals have developed rapidly, this vision is gaining new technical embodiment. At last year's Bund Conference, Ant released GPASS. Over the past year, the team explored with multiple terminal partners services including glasses payment, city tour guides, shared bike-sharing, and parking fees.

In this process, challenges became apparent: repeated chip and system adaptation, difficulty in unified hardware capability invocation, fragmented service ecosystems, and identity and authorization risks during agent execution. This upgrade from GPASS to "Lingyinx" signals that Ant hopes to move further from application and service connection downward to the foundational layer for AI terminal agent execution.

Lingyinx has built an agent execution framework covering interaction, orchestration, policy, execution, tools, and memory, and connects downward to mobile phones, AI glasses, earbuds, and other multimodal terminals.

Users can express needs through natural language and multimodal methods, with the system combining context to understand intent and coordinating devices, agents, and services within clearly authorized scopes to complete tasks.

For developers, Lingyinx provides unified abstraction of hardware capabilities such as camera, display, sensors, and temple interactions, reducing repetitive adaptation across different chips, systems, and devices, allowing developers to focus more on applications and services themselves.

If AI models give machines "intelligence" and agents give them "action capability," what AI terminals need to solve is how to truly bring this action capability into the real world.

This is also leading to new labor division in AI's new economic infrastructure: on one hand, there is the need to solve the "trust problem" of agents entering the commercial world; on the other, there is the "connection problem" of agents entering the real world. The former concerns identity, authorization, and behavioral trustworthiness; the latter concerns how agents understand terminals, invoke devices, and connect services.

Lingyinx hopes to become foundational infrastructure connecting agents to terminals, and terminals to services, allowing hardware from different manufacturers to share agent capabilities, and allowing services developed by different developers to run across terminals.

Security capabilities are also built into Lingyinx's execution chain. The system verifies trusted identity starting from device access, isolates sensitive data locally, and establishes trusted pathways for cross-device connections; when an agent invokes tools or executes tasks, the system performs identity authentication and authorization, with related operations remaining traceable.

Additionally, Ant has partnered with China's Institute of Information and Communications Technology's TAIC Terminal Laboratory to advance testing laboratory construction and security standard-setting, providing support from both technical evaluation and standards specification perspectives for AI terminals to achieve human-centered, secure, and reliable interaction.

Through built-in security mechanisms combined with testing and standard-setting, Lingyinx aims to enable AI terminals to gain "autonomous action" capability while ensuring execution processes have clear boundaries: who is invoking, what is being invoked, what authorization was granted, and how to trace if issues arise.

Whether AI terminals ultimately scale depends not just on one manufacturer's products, but on whether chips, hardware, operating systems, agents, and application services can form synergy. Lingyinx will connect chip partners, terminal manufacturers, and developers through an open platform.

For hardware manufacturers, a unified capability access layer reduces adaptation and development costs; for developers, Lingyinx provides tools including process orchestration, interface components, hardware plugins, and application publishing, supporting application capability reuse across different terminals; for users, lightweight applications can be naturally invoked based on scenarios and discarded after use, reducing complex operations on small-screen or screenless devices.

At the launch, Lingyinx demonstrated multiple generations of reference designs and partner standard solutions, exploring terminal experiences under dual-eye full color, iris eye-tracking, and diverse chip architectures. During the Bund Conference, these capabilities entered real-world scenarios such as Yuyuan Bazaar and the BFC Bund Financial Center, where users can experience services such as city tour guides through AI glasses.

Zhou Zhifeng stated that the AI terminal industry does not need another closed system, but a path that more partners can participate in. In the future, Lingyinx will continue to advance multi-chip adaptation, terminal capability opening, and developer ecosystem building, jointly exploring new interactions and services for AI terminals with industry partners.

From "GPASS" to "Lingyinx," what Ant is doing is not recreating a hardware brand, but attempting to consolidate capabilities including identity, security, payment, agents, and end-cloud coordination into open infrastructure. Like APASS establishing commercial trust for agents, Lingyinx is more focused on enabling agents to truly connect real-world terminals and services.

One addressing "trust," one addressing "connection"—together they form the infrastructure agents need to move from the digital world into real business and real life.