Chinese AI startup StepFun, backed by Tencent, launched the STEPX Neo on October 13, a smartphone designed around an integrated AI agent rather than traditional applications. The device runs Step AOS 1.0, an operating system built from scratch to support autonomous task execution, and features an agent named Amoo that manages user interactions through natural language commands. Pre-orders opened at 3,999 yuan (approximately $560), including 12 months of access to 100 billion tokens per month.
Step AOS 1.0 differs from standard assistants like Siri or Gemini by embedding long-term memory, task scheduling, and security controls into the base layer, utilizing Model Context Protocol (MCP) to break down phone functions into modular blocks. This architecture allows Amoo to execute complex tasks across multiple apps without manual intervention, while simple commands are processed locally on-device. A secondary MiniLED screen on the back displays real-time agent activity.
The launch follows ByteDance’s Doubao phone agent facing restrictions from major Chinese apps in December. To mitigate similar conflicts, nearly 100 partners, including Alipay, Tencent, Ctrip, Didi, and Meituan, have signed on to integrate with Amoo. StepFun initially unveiled the concept in Shanghai on July 13, arguing that conventional smartphone software was not designed for autonomous agents.
StepFun’s strategy addresses the critical interoperability challenges that hindered previous AI-agent initiatives in China’s walled-garden app ecosystem. By securing partnerships with dominant platforms like Alipay and Tencent before launch, the company attempts to bypass the technical and commercial barriers that blocked competitors such as ByteDance’s Doubao. This collaborative approach suggests that successful agent-based hardware requires explicit consent from service providers, shifting the competitive dynamic from unilateral disruption to negotiated integration.
The reliance on Model Context Protocol (MCP) and a hybrid local-cloud processing model highlights the operational risks inherent in this new category. While local processing reduces latency and privacy concerns for simple tasks, complex cross-app workflows depend heavily on cloud connectivity and partner API stability. If key partners restrict access or if the agent misinterprets user intent due to ambiguous natural language inputs, the core value proposition of seamless automation could collapse, exposing users to significant friction and potential security vulnerabilities.
