Xiaomi has moved to make AI agents a mobile- and device-level feature rather than a cloud-only novelty. On March 6 the company began a limited closed beta of Xiaomi miclaw, a phone-based agent built on its MiMo large model and pitched as China’s first mobile implementation of the popular open-source OpenClaw agent architecture.
The product is designed not as a chat assistant but as a system-level agent with four tiers of capability:底层系统 access, personal context understanding, ecosystem connectivity and self-evolution. Xiaomi says miclaw can call apps and IoT devices with user permission, remember user preferences locally, and grow its capabilities through a local memory system — a pitch aimed at turning smartphones into persistent, personal AI controllers for the “people-car-home” ecosystem.
Miclaw’s debut is both symbolic and strategic. The original OpenClaw project — an open, local-first AI agent that gained rapid global popularity late last year — has been dubbed “raising lobsters” because of its red-lobster logo. OpenClaw’s ability to run persistent automation tasks and to link language models to tools ignited intense industry interest and a wave of cloud and device deployments worldwide; Nvidia’s CEO hailed the project as a landmark open-source release.
Xiaomi frames miclaw as the domestic analog that brings agent capabilities to phones and tightly couples them with the company’s Mi Home platform, which already lists over a billion connected devices. That technical tying of agent to device network is Xiaomi’s clear differentiator: an agent that can, with consent, query the status of lights, thermostats and cameras and execute system-level commands on behalf of the user.
The launch also illustrates how big Chinese technology firms are racing to capture the next interface for user attention and control. Tencent, ByteDance and Alibaba have all rolled out ways to deploy OpenClaw or similar agents in the cloud; Tencent reported large in-person demand to install OpenClaw with cloud assistance, while cloud providers have published one-click deployment kits. Xiaomi’s bet, by contrast, is mobile-first and tightly integrated with its hardware and OS ambitions.
That approach raises two tensions. First, the promise of a phone that can orchestrate a home’s devices creates a powerful convenience proposition and a new platform for ecosystem lock-in. Second, security and privacy risks multiply when agents possess system privileges and can orchestrate IoT fleets. China’s industry regulator and security platforms have warned that some OpenClaw deployments, especially with default or improper configurations, can be exposed to attacks and data leaks.
Xiaomi has tried to head off those concerns with technical and policy assurances: miclaw stores conversation histories and permission records locally, prompts users for high‑sensitivity actions, and promises not to use user data to train models. Tencent and cloud providers have introduced sandboxing and layered security schemes to mitigate attack surfaces, but observers say robust operational practices and standards will be essential as agents proliferate.
The miclaw launch also ties into a larger strategic push at Xiaomi. Founder Lei Jun, whose reported personal fortune recently put him back on the global rich list, has earmarked ambitious research spending for core technologies — chips, operating systems and AI — pledging multibillion‑dollar commitments over the next five years. Deploying an agent on billions of phones and linking it to a huge IoT base is a concrete move to solidify Xiaomi’s position in both consumer AI and device-level controls.
For international observers, miclaw is a reminder that the next phase of AI adoption will be decided as much by device makers as by cloud providers. Mobile agents integrated with hardware and home ecosystems can accelerate consumer usage — but they also shift control over data flows and create new regulatory fault lines. Whether miclaw becomes a mass product or a niche experiment will depend on user trust, security hardening, and how well Xiaomi balances local intelligence with safe, auditable behaviour.