Huawei Consumer Business Group CEO He Gang has shared a detailed look at the company’s OpenClaw-based AI agent for HarmonyOS devices, revealing features including self-learning behavior, instant activation, and deep integration with HarmonyOS’s native voice assistant. Pre-orders for the agent, branded “Celia Claw” internationally and “Xiaoyi Claw” in China, are now open for compatible smartphones and tablets.

What He Gang Showed

He Gang posted screenshots on Weibo showing the Claw agent running inside HarmonyOS’s voice assistant interface. According to Huawei Central, the key capabilities include instant access and wakeup (no complex setup required), self-learning behavior that adapts to the user over time, and system-level integration across the HarmonyOS ecosystem.

A Huawei Terminal insider told Wallstreetcn that Xiaoyi Claw’s core is built on HarmonyOS’s foundation, with standout features in security, system-level design, and full-scenario coverage. That last point is significant: Huawei’s HarmonyOS runs across phones, tablets, car systems, PCs, and watches, and Xiaoyi Claw is designed to orchestrate tasks across all of them. A user could issue a command on their phone and have it execute on a PC or tablet in the same ecosystem.

He Gang first teased the product on March 11, posting a screenshot of the beta interface and commenting that he’d been using it for several days and found it “quite useful” and said “it can help me get a lot of things done.” The product has since progressed from internal beta to pre-order availability.

The OEM Angle

Huawei is the largest consumer hardware manufacturer to embed an OpenClaw-derived agent directly into a shipping operating system. Caixin Global reported in mid-March that Xiaomi, Honor, and Huawei were all racing to integrate system-level AI agents, with Xiaomi launching limited testing of its own “MiClaw” on March 6. But Huawei’s integration goes deeper than most: rather than wrapping OpenClaw as a standalone app, the agent is woven into HarmonyOS’s voice assistant and can leverage the operating system’s cross-device handoff capabilities.

According to DQ India, Huawei’s Xiaoyi Claw is currently in beta testing alongside similar efforts from Tencent Cloud (WorkBuddy), Alibaba (CoPaw), and Baidu (ERNIE-powered enterprise agents). The difference is that Huawei controls both the hardware and operating system, giving it tighter integration than cloud-only deployments.

Security as a Selling Point

Huawei is positioning data security as a core differentiator. The Huawei Terminal insider emphasized to Wallstreetcn that “for an AI Agent, security is especially important,” a pointed reference to the security concerns that have dogged OpenClaw deployments on other platforms.

That framing is deliberate. On March 10, China’s National Computer Network Emergency Response Center (CNCERT/CC) issued a security advisory flagging risks with OpenClaw’s default security configurations, including fragile permissions and exposure to full system control if breached. By building on HarmonyOS’s sandboxed architecture rather than running OpenClaw with raw system permissions, Huawei can argue its implementation sidesteps the vulnerabilities the government flagged.

Why It Matters

The move from open-source framework to OEM-embedded product marks a shift in how OpenClaw reaches users. Most OpenClaw deployments to date have required technical setup: installing dependencies, configuring API keys, managing permissions. Huawei’s approach eliminates that friction entirely, putting an AI agent in the hands of consumers who have never heard of OpenClaw and never will.

For the broader agent ecosystem, the question is whether OEM integration accelerates adoption or fragments it. Huawei’s version runs within a walled HarmonyOS garden, leveraging proprietary cross-device features that don’t exist on other platforms. Xiaomi’s MiClaw and Tecno’s EllaClaw are taking similar paths with their own OS integrations. If every major OEM ships a proprietary fork of OpenClaw optimized for its own ecosystem, the open-source framework’s advantage starts looking less like interoperability and more like a reference implementation that everyone customizes beyond recognition.