How HarmonyOS 7's Agent-First Design Could Reshape the Crypto Wallet Experience

The OS as an Agent Broker

When HarmonyOS 7 debuted at HDC 2026, most coverage focused on performance gains and new features. But the deeper shift is architectural: the system is being rebuilt around AI agents. Instead of forcing users to open an app and navigate menus, the OS now tries to understand intent and orchestrate capabilities across apps. For the crypto world, that's a potentially transformative idea. Imagine telling your phone, "Move my ETH to my cold wallet and pay the gas fee with USDC," and having the system handle the multi-step process without you bouncing between five apps.

This isn't just a nicer voice assistant. It's a fundamental change in how applications expose themselves. Apps become schedulable services, registering capabilities that the system can call. For a crypto wallet, that means exposing functions like "sign transaction," "check balance," or "swap token" as structured, callable intents.

Six Layers, One Goal: Intent-Driven Execution

The architecture behind this is HMAF 2.0, the upgraded HarmonyOS agent framework. It's built in six layers:

• Xiaoyi – the system-level assistant that takes user requests.
• HMAF 2.0 – decomposes intent into tasks and manages communication between agents.
• AI Foundation – includes openPangu 2.0, a 505B-parameter open-source model, and a 30B on-device model.
• System Services – Ark engine, StarShield security, and the cross-device link.
• Developer Tools – DevEco Code and DevEco CLI.
• Scenarios – like spatial computing.

For crypto developers, the security layer matters. StarShield uses on-device AI to detect scams in seconds, and it's already blocked 3.47 million potential frauds. That kind of proactive protection could be a selling point for a crypto wallet, where phishing and malicious dApps are constant threats.

Code That Makes Your App Schedulable

To plug into this system, a wallet would register an agent extension. The code is straightforward. You declare capabilities with a schema, and the system matches user intent to your function.

@agentService.AgentExtension
export default class WalletAgent extends agentService.AgentExtension {
declareCapabilities() {
return [{ id: 'wallet.send', description: 'Send tokens', inputSchema: { type: 'object', properties: { token: { type: 'string' }, amount: { type: 'string' }, to: { type: 'string' } }, required: ['token', 'amount', 'to'] } }];
}
async onInvoke(task) {
const { token, amount, to } = task.arguments;
// execute on-chain transaction
return { status: 'success', txHash };
}
}

That's a simplified version, but the idea is clear. Your app becomes a service that the OS can call with structured data, not just a voice command. The system might break down a user's request into sub-tasks, calling your wallet's "send" capability, a calendar agent to schedule a reminder, and a search agent to verify the recipient's address.

Tools of the Trade: DevEco Code and DevEco CLI

Huawei is taking a dual-track approach to developer tools. DevEco Code is an AI-powered IDE that can plan, write, and debug code autonomously. It's like having a senior dev pair programming with you. DevEco CLI, on the other hand, is a command-line interface that exposes HarmonyOS's build, run, and debug capabilities as commands. This lets teams plug into their existing AI agents, like Claude or Cursor.

For a crypto startup, this means you don't have to abandon your current workflow. If you're already using a custom agent, you can just point it at DevEco CLI and start building for HarmonyOS. If you're greenfield, DevEco Code can jumpstart your project, generating boilerplate for features like multi-device adaptation.

But there's a catch. DevEco Code doesn't support Linux yet, which could be a pain for server-side or CI environments. And there's a bigger issue: the AI models don't have enough ArkTS training data. For Swift or Kotlin, AI-generated code works well out of the box. For ArkTS, expect around 15-20% of generated code to need manual fixes. That's a significant tax on productivity, especially for complex crypto logic.

Why Crypto Apps Need Multi-Device Support

HarmonyOS runs on phones, tablets, cars, watches, and TVs. A crypto wallet on your watch could let you approve transactions without pulling out your phone. A DeFi dashboard on your car's infotainment screen? Maybe not ideal, but the point is that your app can seamlessly move across devices.

Huawei's distributed soft bus handles cross-device connections at the OS level. That's a big deal. On Android, you'd have to stitch together different protocols. On iOS, you're locked into Apple's ecosystem. HarmonyOS lets you call a function on another device almost as if it were local.

import { distributedData } from '@kit.ArkData';
await distributedData.applyDeviceProperty('deviceId', { type: 'video', uri: 'datashare://video/live' });

For a crypto app, this could mean starting a transaction on your phone and finishing it on your tablet, or getting a push notification on your watch to confirm a trade. The UX possibilities are real, but they require your app to be built with cross-device in mind from day one.

The Skill Ecosystem: Ready-Made Building Blocks

Huawei offers over 70 pre-built Skills for common tasks, from multi-device adaptation to specific development patterns. These are like plugins that your AI agent can use. For example, there's a Skill for concurrency safety that automatically adds @Sendable annotations to your shared state classes, preventing data races when passing objects across threads.

For crypto apps, concurrency is critical. Transactions, price feeds, and wallet balances update in real-time, and a race condition could lead to incorrect balance displays or even double-spends. Using a Skill that handles this automatically is a no-brainer.

The Skill format follows an open standard called SKILL.md, which Apple also adopts. That means skills written for HarmonyOS could potentially be ported to Xcode, and vice versa. This "Skill-as-Code" approach is becoming an industry norm.

What This Means for Crypto Developers

HarmonyOS is still young. The ecosystem has 400,000 apps, but only 23,000 are native. That's a gap, but it's also an opportunity. Early movers can get ahead of the curve.

If you're building a crypto app, consider these steps:

• Start with DevEco Code if you're new to HarmonyOS. It can shoulder the initial learning curve.
• Use DevEco CLI if you have an existing agent pipeline. It's less disruptive.
• Explore the 70+ Skills to solve common problems without reinventing the wheel.
• Contribute to community knowledge packs like harmonyos-ai-skill, which fill the ArkTS training data gap.

The bottom line: HarmonyOS 7 is betting that the future of operating systems is agentic. For crypto, that could mean a more intuitive, secure, and cross-device experience. The tools are still maturing, but the direction is clear. Now is the time to experiment, before the ecosystem gets crowded.