Smart home technology is evolving from merely "understanding a command" to "remembering a household."
On June 18th, Xiaomi Corp. officially unveiled its "Whole-Home Smart AI Open-Source Solution," Xiaomi Miloco 2.0.
This solution centers on Xiaomi's self-developed MiMo large model as its intelligent core, building upon last year's Miloco 1.0 with upgrades to interaction methods, product features, and notably, a memory system.
Miloco 2.0 primarily integrates via Agent plugins into OpenClaw, supporting macOS, Linux, and Windows systems.
The most significant change is Xiaomi's first-time introduction of a Home Memory AI system.
Previously, the core capability of smart homes was primarily device automation. Users would pre-set rules, and the system would execute them, such as turning on the living room lights at 7 PM, activating the air conditioner when the temperature exceeds 28°C, or switching off lights upon a voice command.
However, this type of smart home essentially remains "rule-based control." It can execute clear instructions but struggles to comprehend the long-formed living habits of a household and cannot continuously adapt services based on the preferences of different family members.
Miloco 2.0 aims to address precisely this capability gap.
According to Xiaomi Corp., Miloco 2.0 can memorize the identities, preferences, schedules, and habits of different household members, building a long-term memory bank for the "home" as a unit. It consolidates observed patterns daily, archiving recurring behaviors into long-term profiles. For instance, the system might notice a family member returning home late and proactively offer a message of care for working overtime.
To support this home memory, Miloco 2.0 utilizes Mi Home cameras as a multi-modal perception entry point, combining microphones, Mi Home devices, and large model capabilities for continuous understanding of the home environment. The system can synthesize information like facial features and body shape to identify household members. If recognition is temporarily impossible, the individual enters a "stranger pool" until confirmed by the user for registration.
This points to a core challenge for large AI models entering home environments: without long-term memory, they remain confined to one-off Q&A and single-instruction execution, which is fundamentally similar to the previous "rule-based control" of smart homes.
From a broader technological trend perspective, Miloco 2.0's significance lies in providing a case study for observing large models entering the physical world.
Historically, large models have primarily operated within chat interfaces, search bars, and office software, with interactions mostly confined to screens.
However, the home is a real physical environment. For AI to be effective here, it must continuously perceive the surroundings, recognize household members, understand behavioral changes, and, when necessary, coordinate real-world devices.
In this context, home memory is precisely the capability that large models must develop as they step into the physical world.
Comments