StepFun’s StepX Neo puts an AI agent at the centre of a smartphone

StepFun’s new StepX Neo phone pairs an AI agent that can act across supported apps with a separate, low-power processor for always-on audio sensing. It is a concrete attempt to make an AI agent part of the phone’s operating system, not just another app waiting to be opened.

StepFun Watch analysis

What happened

The StepX Neo combines StepFun’s Step AOS software and Step Amoo agent with a Snapdragon processor and Syntiant’s NDP120 neural processor, according to Jon Peddie Research. Six microphones feed the dedicated processor, which is described as handling continuous audio tasks while using less power than repeatedly waking the main processor.

Step Amoo can use software interfaces to request actions in supported services, including messaging and ride booking. The report says the initial ecosystem includes Chinese services such as Alipay, Didi, Meituan, JD.com and Amap. The Pioneer Edition is listed at ¥3,999, with 16 GB of RAM and 256 GB of storage. Its purchase includes 12 months of model access, with an allowance of 10 billion tokens per month; StepFun has not disclosed the price after that period.

Why it matters

The interesting design choice is the split between sensing and heavier computing. A small dedicated processor can listen for selected audio events, while more demanding requests go to the phone’s main processor or cloud models. That could make an always-available assistant more practical on battery life, though the report’s figures do not establish how the complete phone performs in everyday use.

An agent that can reach into other apps also makes permissions a central product question. StepFun says selected sensitive actions, including payments and file deletion, require confirmation. Whether users can understand and control the agent’s access will matter as much as how many tasks it can perform.

Our read

This is a notable phone design, not proof that the agent-phone future has arrived. StepFun is joining agent software, app-level actions and specialised low-power AI hardware in one device. The useful test now is whether the integration feels reliable and controllable outside a carefully chosen demonstration. An AI assistant with six microphones is only charming if it knows when to stop listening.

What to watch

• Which apps and agent actions work at launch, and whether the list expands beyond the initial ecosystem.
• How the phone handles permissions, confirmations and visibility into what the agent is doing.
• What model access costs after the included 12 months.
• Whether continuous sensing delivers the claimed power savings in normal use.

Discussion spark: Would you trust a phone agent to act across your apps if sensitive actions still required confirmation, or should its access be limited by default?

not affiliated with or endorsed by StepFun