Xiaomi's mobile AI Agent "Xiaomi miclaw" begins limited closed beta testing.

Xiaomi miclaw, an AI interaction testing product built on Xiaomi's MiMo large-scale model, began its limited closed beta test today, by invitation only, with no public recruitment. Running as a system application, Xiaomi miclaw encapsulates over 50 system-level tools and ecosystem services, employing an inference-execution loop engine where the model autonomously determines the tool invocation order and parameters. This product focuses on verifying the execution capabilities of the large-scale model within Xiaomi's "Human, Vehicle, and Home Ecosystem," featuring a four-layer architecture: system-level capabilities, personal contextual understanding, ecosystem interconnection, and self-evolution. Regarding ecosystem interconnection, Xiaomi miclaw implements a complete Mi Home protocol client, allowing control of IoT devices connected to Mi Home with user authorization, and supports the MCP protocol and third-party application access SDKs. In terms of self-evolution, the product possesses four meta-capabilities: file-level memory, sub-agent creation, MCP service configuration, and sandbox script execution, enabling it to autonomously build a memory system and expand its tool scope. Xiaomi notes that the product is still under optimization in terms of stability, power consumption, and success rate in complex scenarios; users are advised to back up their data beforehand and test in a controlled environment.