AI Agent Phone: StepFun’s STEPX Neo Explained13 min read2,549 words

AI Agent Phone: StepFun’s STEPX Neo Explained

StepFun’s STEPX Neo puts an AI agent inside the phone’s operating system. See how Amoo works, what it costs, and what it could mean for app platforms.

AI agent phoneStepFun STEPX NeoAmoo AI agentAI smartphone operating systemAI and crypto adoption
AI Agent Phone: StepFun’s STEPX Neo Explained

StepFun's STEPX Neo Puts an AI Agent at the Center of the Phone

Smartphones have spent years sprinkling AI into familiar apps. StepFun is trying something more radical: instead of treating AI like just another app, it wants it to sit underneath everything - coordinating the phone's services on your behalf.

The Chinese startup opened pre-orders for the STEPX Neo on October 13 for 3,999 yuan (about $560). It runs Step AOS 1.0, an operating system built around an AI agent called Amoo. Step AOS is designed to interpret a goal and complete tasks across supported services. The price includes a year of AI access, with a stated allowance of 100 billion tokens per month. StepFun has not said what it will cost after that first year.

That framing matters, because this isn't only a hardware launch. It's a test of whether people will trust an agent to act across services - and whether app companies will play along once the operating system starts making decisions for the user. And while that sounds like "phone news," the same issues show up in crypto wallets and financial apps: convenience has to be weighed against permissions, security, and control.

StepFun's materials don't point to a blockchain feature in the STEPX Neo. If crypto is relevant here, it's mainly as part of the bigger shift toward agent-driven interfaces.

What StepFun Launched

StepFun introduced the STEPX brand, its operating system, and the Amoo agent at an event in Shanghai on July 13. Pre-orders for the STEPX Neo opened on October 13 through JD.com, Tmall, and Douyin, with an initial sales event planned for October 16. The reported price is 3,999 yuan (about $560).

StepFun's main claim isn't just that Amoo can answer questions. The company says the phone is set up so the agent can take a user's goal and carry out steps across supported functions and services. In a typical ride-hailing flow, you open an app, enter a destination, pick a vehicle, and confirm. With an agent-based approach, you describe the destination and delegate some or all of those steps.

On paper, that doesn't sound huge. But once you move beyond one app, it changes how daily phone use works. Today, you bounce between interfaces and follow each app's steps. In an agent model, you set an outcome and let software handle the actions needed to reach it. The interface shifts from navigation to delegation.

The price includes an AI allowance

The STEPX Neo's price includes 12 months of the STEPX Token Plan, which StepFun says provides 100 billion tokens per month. Tokens are the units AI providers use to process text and other inputs. They're not a direct measure of how many tasks a person can complete. The real value of the allowance depends on how the service counts usage, which models are used, and what limits apply to different requests.

StepFun hasn't said what access will cost after the first year. That matters because owning an agent-based device isn't just a one-time purchase - ongoing inference costs typically depend on usage and how much processing happens in the cloud. Until StepFun publishes renewal pricing and usage rules, buyers can't estimate the total cost of ownership over multiple years.

The reported launch details come from Decrypt's coverage of the STEPX Neo. The included token plan should be treated as a launch-year benefit, not a guarantee of a permanent service level.

How Amoo and Step AOS Are Designed to Work

StepFun positions Step AOS 1.0 as an operating system built around an agent, not as a conventional phone OS with an assistant bolted on later. The company describes it as combining Android, Linux, and real-time operating system components. StepFun says the base layer includes long-term memory, task scheduling, and security controls.

The architecture matters because an agent needs more than a language model. It needs a way to figure out what actions are available, call the right service, track what's happening, and respond when something fails. If a step doesn't work, the system has to decide whether to retry, ask the user for clarification, or stop. That's orchestration - just as much as it's model performance.

Step AOS reportedly breaks up functions like messaging, files, apps, and settings into smaller units that the agent can call directly. It uses the Model Context Protocol (MCP), a standard for connecting AI systems to external tools and software. The basic idea is to give the agent a set of defined routes into services, rather than forcing it to guess how to operate every screen.

Two interfaces, with different levels of delegation

The STEPX Neo still includes a familiar app grid, along with a separate space for Amoo. That kind of split layout is a reasonable transition: people can stick with manual control for tasks that feel sensitive, weird, or just faster to do themselves, and delegate routine work when the agent is likely to help.

StepFun also describes two kinds of memory: one for user facts, habits, and preferences, and another for knowledge about carrying out tasks. Separating those could help the agent personalize suggestions while also improving repeat workflows. But memory also creates governance questions: what gets stored, how long it sticks around, how people can review or delete it, and whether it's processed locally or in the cloud.

You can't really treat those concerns as "later." Persistent memory can make an agent more useful, but it also builds a fuller picture of someone's routines. The product will need controls people can understand - more than just a privacy policy that reads complete on paper.

Local actions and cloud reasoning

According to StepFun, simpler actions like setting an alarm can run on a smaller model inside the phone, with some features available offline. More complex requests that span multiple apps can be sent to a larger cloud model. This hybrid approach has a practical upside: local processing can reduce latency and cut down on sending routine requests to remote services, while cloud models can handle tasks that need more horsepower.

It also brings trade-offs. Cloud-based actions require connectivity and may involve sending personal context off-device. Local processing can be faster and more resilient, but phone hardware limits what it can do and how much it can handle. Users will need to understand which path a task takes - especially for things involving messages, payments, identity, or financial data.

A MiniLED screen on the back of the STEPX Neo is meant to show what Amoo is doing. That could be helpful if it makes the agent's activity visible. But a light or indicator isn't the same as meaningful confirmations and clear permissions. It can show that something is happening; it doesn't automatically explain what data is being shared or how to undo an action if something goes wrong.

Why App-Company Participation Matters

StepFun says nearly 100 companies and products have signed on to Amoo, including Alipay, Tencent, Ctrip, Didi, and Meituan. The model is that supported services handle actions like payments, travel bookings, and content creation on their own platforms, while Amoo coordinates the user's request.

This approach is aimed at a known weakness in agent-based phone control: screen automation tends to be fragile. If an agent reads the screen and simulates taps, it can break when buttons move, pop-ups appear, or an app updates its layout. There's also the issue of whether a service provider can tell the difference between a real user action and automated behavior.

Direct integrations can help because the agent has defined capabilities instead of relying on visual guesswork. But that still doesn't automatically make the system safe. Integrations still require clear permissions, limits, error handling, and accountability. A payment action shouldn't be treated like opening a menu. The system has to distinguish between finding information and authorizing a transaction that changes something real.

The platform-control question

Integration choices aren't just technical - they're also commercial. If a company's service is accessible through an agent, users may rely less on opening that company's app directly. That can shift who controls discovery, customer relationships, and the moment where decisions are made.

There's also a real-world example from China. According to the source report, ByteDance's Doubao phone agent ran into blocks from major apps in December after it used screen-reading and simulated clicks. The tension is straightforward: users often want automation, while service providers may not want a third party operating their apps or redirecting attention away from their own interfaces.

StepFun's partner-based approach may reduce some of that friction. Still, it makes the agent dependent on participating services. If a partner changes terms, removes an integration, or limits access, a workflow can disappear. "Nearly 100 partners" is a big headline, but the number alone doesn't tell you how reliably each integration works, what tasks are supported, or whether access will remain available over time.

For users, the practical test isn't the partner count. It's whether common tasks work consistently, whether the agent asks before actions with real consequences, and whether you can see what it did. For service providers, the key question is whether the integration gives them enough control over security, consent, and the user experience.

What the Phone Could Mean for AI and Crypto Adoption

The STEPX Neo is an AI phone, not a crypto phone. The reported launch details don't mention a blockchain, crypto wallet, token, or decentralized application as part of the product. Any link to crypto is therefore more about where agent interfaces could go than about what StepFun is selling right now.

The bigger idea is that agents could become a new front door for financial services. In crypto, a future agent might help check a wallet balance, compare network fees, or draft a transaction. But drafting and signing are not the same thing. Even if an agent can explain options or assemble a draft, users should still have clear, deliberate control over the private key and final approval.

That distinction is central to safer adoption. Crypto transactions are often difficult or impossible to reverse. If an agent misunderstands a request, picks the wrong network, or interacts with a malicious service, the user can be left holding the loss. Agent-driven convenience can also hide the moment when a mistake should be noticed. Convenience without explicit authorization isn't really an upgrade.

The same trust requirements apply to mainstream finance

StepFun's financial-service partnerships, including Alipay, make its ambitions relevant even without a crypto component. If an agent coordinates payments or bookings, it still has to handle identity, consent, and confirmation in ways users can follow. That applies whether the payment ends up as a bank transfer, card charge, or a blockchain transaction.

For crypto platforms and wallet developers, the lesson isn't that an AI agent should take over signing. It's that integrations should expose narrow actions, with permissions matched to risk. Reading a balance is different from sending funds. Searching for a flight is different from buying a ticket. An agent's capabilities should reflect those differences.

A responsible design would make the boundary visible:

  • Read-only actions can usually be delegated with lower risk.
  • Reversible actions need clear status and an easy way to cancel.
  • Financial or irreversible actions should require a specific confirmation that names the asset, recipient, amount, and relevant network or service.
  • Sensitive data access should be limited to what the task requires, with a clear way to revoke access.

Those are design principles, not claims about what STEPX Neo supports today. StepFun says the operating system includes security controls, but the available launch information doesn't explain the details of how those controls work. That gap should be treated as an open question, not something to fill in after the fact.

What to watch after launch

There are three practical tests for the STEPX Neo. First, does the agent complete everyday tasks reliably across its partner services? Demos can look great, but sustained use will reveal edge cases, app changes, and unclear requests.

Second, is the operating model financially sustainable? The phone includes 12 months of token access, but StepFun hasn't disclosed the post-year price. If generous usage is expensive to deliver, the company may need to set limits, introduce tiers, or subsidize access through other parts of its business. Each approach affects the user experience.

Third, do users understand what Amoo is doing? A capable agent isn't automatically a trustworthy one. People need visibility into what information it used, which services it contacted, what it changed, and what still requires approval. In real production systems, that operational clarity matters more than a polished demo.

Conclusion

The STEPX Neo is an early test of a different phone model - one where the operating system is built to let an AI agent coordinate tasks across services. StepFun's approach combines on-device and cloud models, a dedicated agent interface, and a partner network that includes major Chinese platforms.

The key question isn't whether an AI agent can complete a task in a launch demo. The real question is whether it can do so reliably and transparently at a sustainable cost - and whether users keep meaningful control when it matters. Those standards apply to everyday bookings and payments, and they become even more important if agent interfaces expand into crypto wallets or other irreversible financial actions.

If you're evaluating the STEPX Neo, focus on real-world reliability, the renewal price after the included year, and clear explanations of permissions and confirmations. For crypto product teams, the takeaway is simple: make delegation genuinely useful, but keep authorization explicit. Read more about StepFun's launch and product details before treating the device's AI claims as a substitute for hands-on testing.

Frequently Asked Questions

The STEPX Neo is a smartphone from Chinese AI startup StepFun. It runs Step AOS 1.0, an operating system designed around Amoo, a built-in AI agent intended to carry out tasks across supported phone functions and services.
StepFun opened pre-orders at 3,999 yuan, about $560. The price includes 12 months of the STEPX Token Plan, which the company says provides 100 billion tokens per month. StepFun has not announced the service price after the first year.
Amoo is StepFun’s built-in AI agent. The company says it can use phone functions and supported services to complete tasks. Simpler requests may run on smaller models on the device, while more complex work can be sent to a larger cloud model.
The launch information does not identify a blockchain, crypto wallet, or cryptocurrency feature in the STEPX Neo. Its connection to crypto is indirect: agent-based interfaces raise relevant questions about permissions, transaction approvals, and user control for future financial applications.
Direct service integrations can be more reliable than an agent that operates apps by reading screens and simulating taps. They can also make the agent dependent on partner access. The quality of the network will depend on which tasks work, how reliably they work, and what controls users and service providers have.

Related articles