1. What is the StepFun AI phone?
StepFun, a Chinese artificial intelligence startup backed by Tencent, opened pre-orders on October 13 for the STEPX Neo, a smartphone priced at 3,999 yuan, or about $560. What sets the device apart from conventional handsets is that its AI agent is built into the operating system itself, rather than installed as an app on top of it. A first sales event followed on October 16.
An operating system is the core software that runs a phone, in the way iOS runs an iPhone and Android runs most other handsets. By writing its own, StepFun is attempting something more ambitious than adding a chatbot to an existing platform. It wants the agent, not the grid of apps, to be the primary way a person gets things done on the device.
The company described the shift to its launch audience in direct terms. In the agent era, it said, AI is no longer an add-on to a smartphone but the product itself.
2. How does an agent-first phone differ from today's assistants?
The practical difference shows up in everyday tasks. On a typical phone, a person who wants a ride opens a ride-hailing app, types an address and chooses a car. On the STEPX Neo, the user states the destination and the agent takes care of the remaining steps. An agent, in this context, is software that is given a goal and carries it out on its own, instead of waiting for instructions at every stage.
Existing voice assistants such as Siri and Gemini live inside iOS and Android respectively. They sit on top of systems designed around apps and taps. StepFun argues that its Step AOS 1.0 operating system was built from scratch for agents, with long-term memory, task scheduling and security controls placed in the base layer. The system is described as a blend of Android, Linux and real-time operating system components.
That design choice reflects a view StepFun first voiced in July, when it showed the agent, the operating system and the STEPX brand in Shanghai on July 13, without a price or a sales date. At that event the company said conventional smartphone software was never built for autonomous task execution.
3. How does Step AOS 1.0 let the agent control the phone?
Step AOS divides phone functions such as messaging, apps, files and settings into small building blocks that the agent can call on directly. It uses the Model Context Protocol, known as MCP, a shared standard that allows AI programs to plug into other software. Because the blocks are exposed in a standardized way, the agent does not have to imitate a human tapping through screens.
Users are not forced to abandon the familiar layout. The phone offers two entrances: the traditional app grid and a separate agent space where the assistant operates. That arrangement lets someone fall back on apps whenever they prefer to, while moving routine multi-step tasks to the agent.
This also differs from the screen-reading method used by ByteDance's Doubao phone agent, in which software looks at what is displayed and simulates clicks. The STEPX Neo instead builds on agreements with partner services.
4. What is Amoo and how does it remember things?
The agent is named Amoo. It keeps two separate memory stores: one holds the user's facts, habits and preferences, and the other records what the agent has learned about how to carry out tasks. The first lets it personalize requests, while the second lets it improve at completing them over time.
Processing is split between the device and the cloud. Simple commands, such as setting an alarm, run on a small on-device model, and some of them work offline. Harder requests that span several apps are sent to a larger cloud model. This division keeps quick tasks fast and private while reserving heavier computation for the situations that demand it.
A second screen using MiniLED technology on the back of the phone shows what Amoo is doing at any moment. The display gives the user a way to follow the agent's progress without opening the main screen, which addresses a common concern about agents acting out of sight.
5. Which companies have signed on as partners?
Nearly 100 companies and products have signed on to work with Amoo. They include Alipay, Tencent, Ctrip, Didi and Meituan, which together cover payments, travel, ride-hailing and local services. These are the kinds of platforms through which many users in China already handle daily errands.
Under the arrangement, payments, travel bookings and content creation are still completed by each partner's own platform. The agent coordinates the request, but the transaction itself happens inside the partner's system, where its own security and account rules apply.
The partnerships carry added weight because of a precedent. ByteDance's Doubao phone agent, which relied on reading the screen and simulating clicks, drew blocks from major Chinese apps in December. StepFun's decision to sign agreements with partners is presented as a response to that episode.
6. Why does it matter that the agent sits in the operating system?
Placing the agent in the operating system changes what it is allowed to do. An app-based assistant works within the limits that the host platform sets: it can usually act only when the system grants specific permissions, and it often has to ask the user to confirm each step. An agent embedded in the operating system can be given deeper access to messaging, files and settings from the start, because the platform was designed with that access in mind.
Memory, scheduling and security are the three capabilities StepFun says it placed in the base layer. Each addresses a familiar weakness of add-on assistants. Without memory, an assistant treats every request as new. Without scheduling, it cannot carry a task through several stages or return to it later. Without built-in security controls, giving software permission to act on a person's behalf creates obvious risk, since the same access that lets an agent book a trip could be misused if it were poorly protected.
The trade-off is that a new operating system must persuade developers and partners to support it. Handset makers that rely on Android benefit from a vast catalog of existing apps, and a platform with its own architecture has to build its ecosystem from nearly 100 launch partners rather than millions of apps. That is why the partner list, rather than the hardware, may prove the more important part of the launch.
7. What are tokens, and why does the plan count them?
The bundled plan is measured in tokens, the unit in which AI services count and bill their work. A token is a small piece of text, and every question a user asks and every answer the model produces consumes some. An agent that handles multi-step requests uses considerably more of them than a simple chatbot, because it may read information, plan, call several services and check results before reporting back.
Setting the allowance at 100 billion tokens a month, for 12 months, is therefore a statement about how StepFun expects the device to be used: continuously and with heavy reliance on the agent. It also shifts the cost of running the AI from the user's monthly bill to the up-front hardware price, at least for the first year.
What happens afterward remains unspecified, and that uncertainty is the main commercial unknown. If the company offers a renewal at a modest price, the device could remain attractive. If it does not, buyers may face the choice of paying for continued access or losing the feature that defines the phone.
8. Who is behind StepFun and what models does it build?
StepFun was founded in 2023 by Jiang Daxin, a former Microsoft vice president. The company builds its own large language model, or LLM, which is the kind of AI system trained on large volumes of text to generate and interpret language. Owning the model gives it control over the software that powers the phone's agent.
Its flagship, Step 5 Preview, arrived in September. It has 600 billion parameters, of which 27 billion are active at any one time, a structure that lets a very large model run more efficiently by using only part of its capacity for each request. Step 5 Preview can process up to 1 million tokens at once. Tokens are the chunks of text that AI services count and bill by, so a larger window means the model can consider far more material in a single exchange.
A smaller model, Step 3.5 Flash, was released earlier this year as open-source software under the Apache 2.0 license, which permits wide reuse by developers.
9. What does the 3,999-yuan price include?
The 3,999-yuan price covers 12 months of the STEPX Token Plan, which provides 100 billion tokens a month. Pre-orders were available through JD.com, Tmall and Douyin, three of China's largest e-commerce and short-video channels. Bundling a year of AI access into the hardware price signals that StepFun views the phone as a delivery vehicle for its AI service as much as a piece of electronics.
One open question is the cost after the first year. StepFun has not said what the AI will cost once the included 12 months run out. For buyers, that matters because the agent is central to the device, and an ongoing subscription could change the long-term cost of ownership.
For the broader market, the STEPX Neo is an early test of whether consumers want a phone organized around an agent. Its fortunes will depend on how reliably Amoo completes tasks with partner services, how the company prices its AI after the first year, and whether other app makers keep cooperating with agent-driven requests rather than blocking them.