Xiaomi Miclaw: Xiaomi's AI Agent Exploration Product Enters Closed Beta

Forum topic · 小凯 · 2026-03-06

Summary

Xiaomi has unveiled Xiaomi Miclaw, an AI Agent exploration product built on the MiMo large language model, with small-scale closed beta testing beginning March 6, 2026. The product aims to validate large models' execution capabilities across Xiaomi's Human-Car-Home ecosystem, moving beyond conversational AI toward system-level execution. Miclaw features a four-layer architecture: (1) system-level capabilities including a native system app identity, 50+ system tools, an autonomous reasoning-execution engine, three-tier context memory supporting 20-step complex operations, and prompt caching that cuts token costs by 50-90%; (2) personal context understanding, demonstrated by travel preparation from ticket SMS and detection of duplicate subscription charges, with local storage of chat history; (3) ecosystem connectivity via a full Mi Home protocol client controlling over 1 billion IoT devices, MCP client support, and an open SDK for third-party apps; (4) self-evolution through file-level memory, sub-agent creation, MCP service configuration, and sandboxed Python/JavaScript scripting. Testing is invitation-only; Xiaomi warns against installing on primary devices and recommends data backups.

Xiaomi has released Xiaomi Miclaw, an AI Agent exploration product built on the MiMo large language model. Small-scale closed beta testing begins March 6, 2026.

Product Positioning

Miclaw focuses on validating large model execution capabilities within Xiaomi's "Human-Car-Home" full ecosystem, exploring the path from "conversational ability" to "system-level execution ability."

Four-Layer Capability Architecture

1. System-Level Foundation: Giving AI Hands and Feet

System app identity: Core system application capabilities, not running in a sandbox
50+ system-level tools: Wrapping phone system capabilities and ecosystem services
Reasoning-execution engine: The model autonomously decides call order, with streaming progress updates
Context management: Three-tier intelligent memory management; no losing track across 20-step complex operations
Token optimization: Multi-level prompt caching design, saving 50%-90% token overhead

2. Personal Context: Truly Understanding You

Perceive → Associate → Judge → Act: The dividing line between AI Agents and traditional software
Example - automatic travel preparation:
Example - spending insights:
Data security: Chat history stored locally; only the current conversation is transmitted to the cloud, encrypted

3. Ecosystem Connectivity: From One Phone to the Entire Smart Life

Mi Home ecosystem: Full Mi Home protocol client, capable of controlling over 1 billion devices
Capability translation: Compiles IoT device specs into natural language the model understands
Example - home office mode:
MCP protocol: Full MCP client, supporting thousands of MCP tools on PC
Open SDK: Third-party apps can proactively declare capabilities for the AI to invoke on demand

4. Self-Evolution: Continuous Capability Growth

File-level memory: Creates arbitrary data structures, designs its own memory system
Sub-agent creation: Specialized division of labor with independent prompts and tool whitelists
MCP service configuration: Dynamically connects external capabilities
Script execution: Runs Python/JavaScript in a sandbox

Demo Scenarios

"Bringing a friend home in half an hour" → auto-prepares welcome scene (lights/AC/music)
"Look at food photos in my album, help me lose 3kg next week" → analyzes eating habits and makes a plan
"Wake me at 7am tomorrow, and if I don't get up, make a real fuss"
"Every morning at 8:16, voice-broadcast my schedule plus AI financial news"
"Broadcast gold prices every two minutes"

Beta Information

Closed beta, invitation-only, no public recruitment
Stability, power consumption, and complex-scenario success rates are still being optimized
Not recommended on primary devices; back up your data first

Original source: https://mp.weixin.qq.com/s/TLU6WXkgI-7Ph2ebGQKARg