After Doubao Hits a Wall, STEPX Neo by Leap Star Aims to Connect App Islands with Protocols

On the evening of July 13, 2026, Leap Star launched the world's first large model native intelligent agent smartphone, STEPX Neo. At the launch event, Leap Star's chairman, Yin Qi, candidly admitted that they consulted many terminal friends, and everyone advised them not to touch hardware. They actually wanted to heed this advice but ultimately decided to proceed. This statement reveals the underlying anxiety of large model companies venturing into hardware. Just six months ago, ByteDance's Doubao mobile assistant faced regulatory bans for forcibly crossing the WeChat ecosystem's walls. The two products represent two distinctly different paths for AI smartphones in cross-application operations. Can STEPX Neo truly break the long-standing App ecosystem walls?

Main visual of the launch event for Leap Star's first intelligent agent smartphone, STEPX Neo

Not Just a Parasite in Apps

To understand why Leap Star insists on making smartphones, one must first grasp the funding flow and monetization dilemmas in the large model industry. According to investment reports, in January 2026, Leap Star completed a B+ round of financing exceeding 5 billion yuan, setting a record for the highest single investment in the large model sector in nearly 12 months. Founded in April 2023, the company, led by founder Jiang Daxin, has built a solid technical foundation in the multimodal large model field. Despite having substantial funds, they chose to enter the hardware sector with a long supply chain primarily because large models struggle to monetize in the cloud and urgently need a super entry point on the device side.

Currently, most large model applications exist in the form of independent apps or mini-programs. Whether it is Leap Star's Step model or other manufacturers' large models, once they are parasitic on existing mobile operating systems, they can only be constrained by the permissions and traffic distribution rules set by system vendors. Without mastering system-level entry points, agents will forever remain parasites within apps, unable to truly reshape users' workflows. By personally venturing into smartphone manufacturing, large model companies essentially aim to start from the underlying system, gain control over hardware entry and system-level permissions, and make intelligent agents the core hub of the operating system.

STEPX Neo is equipped with the Step AOS system and features the built-in intelligent agent Amoo. According to reports from the Financial Association, this model is manufactured by Huaqin Technology and features an interactive secondary screen on the back, set to debut at the World Artificial Intelligence Conference on July 17. Specific hardware configurations and pricing have yet to be announced. From the product's form, the design of the interactive secondary screen may be intended to provide a constant display of the intelligent agent's status and a quick interaction entry, further reinforcing its positioning as a native intelligent agent hardware. Leap Star attempts to demonstrate that large models can not only serve as cloud brains but also become super butlers on the device side.

System-Level Permissions Are Not a Universal Key

To assess the tool value of STEPX Neo, one must first revisit the lessons learned from Doubao's wall collision. In December 2025, ByteDance launched the Doubao mobile assistant. According to media reports at the time, the assistant was first launched on the ZTE Nubia M153, priced at 3,499 yuan. The core selling point of this product was its ability to simulate clicks through system-level permissions to achieve cross-application operations. Users only needed to issue commands to Doubao, and it could automatically open other apps, bypassing splash ads and directly accessing information streams or transaction pages.

This unobstructed simulation click technology belongs to an external operation model. Initially, it indeed brought stunning demonstration effects, showing users the potential for cross-application automation. However, it soon hit a wall. In early December 2025, a large number of users reported that the Doubao mobile assistant triggered security controls when operating WeChat, leading to abnormal exits or login failures. Subsequently, Doubao removed this capability.

This was not merely a technical failure but a direct conflict within the commercial ecosystem. The core interests of super apps lie in traffic distribution and ad display. Doubao's simulated clicks directly bypassed these intermediate links, touching WeChat's bottom line. From a technical perspective, simulated clicks forcibly read screen pixels and mimic human finger actions, making this external behavior easily recognizable as abnormal operations by the app's risk control system. App vendors only need to upgrade UI layouts or add verification steps to easily intercept simulated clicks.

Legal experts have also pointed out that AI intelligent agents relying on accessibility services to forcibly cross applications pose risks of unfair competition, making confrontational paths unsustainable. Doubao's wall collision declared the death penalty for relying solely on brute-force cracking routes in the commercial ecosystem. System-level permissions are not a universal key; forcibly prying open someone else's door will ultimately invite tighter lockdowns.

Turning Confrontation into Cooperation with GUI-MCP

Faced with the same cross-application operation challenges, Leap Star has proposed a different solution. STEPX Neo did not choose simulated clicks but instead released the Step Edge device-side model and the GUI-MCP protocol. According to Leap Star's open-source documentation, this protocol adopts a layered dual-stack architecture, supports high privacy modes, keeps original screenshots local, and only uploads semantic summaries to the cloud. The open-source GUI proprietary model has a parameter count of 4B and claims to support local recognition and operations across over 200 apps.

This protocol-based approach fundamentally differs from Doubao's external operation model. Simulated clicks forcibly take over the user interface, while the GUI-MCP protocol attempts to standardize interfaces, encouraging app vendors to willingly cede capabilities. Leap Star hopes to shift from confrontation to cooperation, allowing apps to become skill providers for intelligent agents rather than targets for cracking.

Under the layered dual-stack architecture, the device-side model handles visual screenshots and local operations, while the cloud model is responsible for high-level planning and complex logical reasoning. This end-cloud collaborative design ensures both response speed and reduces cloud computing costs. More importantly, the privacy protection mechanism of keeping original screenshots local and only uploading semantic summaries alleviates users' concerns about data leakage and provides a secure foundation for app vendors to open interfaces.

From a tool evaluation perspective, protocol-based interfaces far outperform simulated clicks in execution efficiency and stability. Through standardized interfaces, intelligent agents can directly invoke core functions of apps without blindly searching for buttons on the screen. This means that when handling complex multi-step tasks, STEPX Neo can theoretically provide a smoother experience. However, all of this hinges on whether app vendors are willing to integrate the GUI-MCP protocol.

The Essence of Breaking Walls Is to Restructure Benefit Distribution

The differences in technical routes ultimately point to the competition within the commercial ecosystem. Leap Star's announced first batch of ecological partners includes Meituan, WPS, Jianying, Ctrip, Gaode, Alipay, Baidu, Didi, and JD.com. These partners cover core scenarios such as food delivery, office work, travel, and payments. Leap Star aims to establish standardized open protocols by attracting these leading applications.

However, the essence of breaking ecological walls is not a technological breakthrough but rather the distribution of benefits. The core value of AI agents lies in de-appification, allowing users to reach task settlement pages directly. This directly threatens the cake of super apps. Without a reasonable commercial sharing mechanism, super apps will continue to counteract agents by modifying code and upgrading verification codes.

In the industry, there is actually another path, where apps actively package their capabilities as Skills or MCPs to open up to agents. This proactive openness can avoid confrontation, but the premise is that app vendors believe the benefits of openness outweigh the losses of a closed moat. Leap Star's first batch of partners are mostly business-oriented applications that need orders and traffic and are willing to try new entry points. However, whether giants with walled gardens, such as Tencent, are willing to adopt this standardized protocol remains unclear without an official response.

Whether Leap Star's ecological alliance strategy can truly convince all giants remains a significant question mark. Without the support of national-level applications like WeChat and Taobao, STEPX Neo's cross-application capabilities will be greatly diminished. Establishing a dual authorization mechanism and a reasonable benefit-sharing model is the only way for AI smartphones to break through ecological walls.

From Parameter Stacking to Workflow Restructuring

As a tool, what impact does STEPX Neo have on the actual workflows of ordinary users? As the ultimate tool carrier, can the intelligent agent attribute of AI smartphones execute complex tasks across application islands? This is the core criterion for judging its tool value. This goes beyond mere stacking of hardware parameters and directly addresses the practical pain points of AI tools.

OmniTools believes that the true tool value lies in seamless integration into workflows. STEPX Neo aims to allow users to no longer frequently switch apps but to complete cross-application multi-step tasks directly through the Amoo intelligent agent. For example, users only need to say, "Help me book a flight to Beijing tomorrow and get a ride to the company," and the intelligent agent can automatically call the interfaces of Ctrip and Didi to complete the booking and ride-hailing. Once this experience is realized, it will fundamentally change the way people interact with their smartphones.

However, there remains a gap between ideals and reality. Currently, STEPX Neo has only been showcased at the launch event, lacking real-world testing data in complex scenarios. In the face of pop-ups, verification codes, payment risk controls, and other extreme situations, the stability of the GUI-MCP protocol still needs to be validated at the WAIC debut. Additionally, hardware supply chain management capabilities are also a shortcoming for large model companies. While Huaqin's OEM can ensure mass production, it lacks the offline channels and after-sales capabilities of traditional smartphone manufacturers.

The launch of STEPX Neo marks the beginning of large model companies transitioning from the cloud to the device side, from the application layer to the system layer. It proposes a different ecological cooperation path with the GUI-MCP protocol, distinct from brute-force cracking. However, whether it can truly break the App ecosystem walls depends not only on the maturity of the technology but also on the restructuring of the benefit distribution mechanism. Until the WAIC's real machine debut, all of this remains a suspense.