Xiaomi officially open sourced Xiaomi Miloco 2.0 on June 18, bringing the AI smart butler into ordinary households.
The industry says this is a functional upgrade for smart homes, but more importantly, the competitive logic of smart homes has changed from competing for the number of devices to competing for AI brain power.
Why is Xiaomi the first to come up with an active AI solution with family memory?
The industry changes behind this are more worth pondering than the product release itself.
In the past ten years, smart homes have been unable to break out of a strange circle: you have to shout commands and tap the screen for the device to move. The essence is still an “advanced remote control”.
The changes brought about by Miloco 2.0 this time are exactly the opposite of this logic.
AI no longer waits for you to give orders. It will proactively sense, judge, and serve.
When you open the door after working overtime until late at night, AI remembers your daily home time and takes the initiative to greet you; when you forget to turn off the heat while frying steak, AI senses the risk through the camera and temperature sensor and sends a reminder 20 seconds in advance; when an elderly person forgets to take medicine, AI combines the family health file and uses the speaker to broadcast reminders on time.
Five years ago, these scenes only existed in “Jarvis” in science fiction movies. Today, Xiaomi has made it an open source solution that can be implemented.
What’s even more interesting is that it is not simply a matter of plugging a large model into a smart home, but a complete set of “perception-decision-execution” closed loop.
Xiaomi summarizes this set of capabilities into four core upgrades: multi-modal perception, active intelligence, continuous tasks, and family memory.
The most subversive one is the function of family memory.
In the same study room, when the male owner sits down to use the computer, the AI will automatically brighten the warm light; when the female owner sits down to write notes, the AI will change to soft neutral light; when it recognizes that the female owner’s sitting posture is wrong, the AI will also combine her history of waist discomfort to instantly remind her to make adjustments.
Such a personalized experience is simply not possible with rule-driven smart homes in the past.
The greatest kindness of technology is that it remembers all your little preferences without you having to repeat settings again and again.
From “what you want the device to do” to “the device knows what you want to do”, this step may seem simple, but it is a key turning point for smart homes to move from “semi-automatic” to “fully automatic”.
Many people only saw the functional innovation of Miloco 2.0, but did not understand why Xiaomi was able to do this.
In essence, this is not the victory of a software solution, but the fruit of the ecosystem of 1.2 billion connected devices.
According to the latest public data, as of June 2026, the number of Xiaomi’s global AIoT device connections has exceeded 1.2 billion, a year-on-year increase of 18.3%.
This number is a unique scale advantage on a global scale.
You can think of Miloco 2.0 as an “AI brain” that requires “eyes, ears, and skin” all over the house to collect information, and Xiaomi’s Mijia ecosystem just covers these sensory entrances to thousands of households.
Cameras provide visual perception, temperature sensors monitor environmental changes, speakers perform voice output, and bracelets collect user health data… All these devices have already completed market education and user penetration. Miloco only needs to connect them and use AI to turn data into services.
Looking at other players, either the device ecology is fragmented, and devices of different brands cannot even be interconnected; or they only make high-end solutions, which are difficult to implement in ordinary families.
The threshold for Xiaomi’s combo is actually much higher than it seems.
What’s more important is that Xiaomi has directly open sourced Miloco 2.0 this time, and also implemented one-click deployment. Ordinary users can experience it as long as they have a computer that can run OpenClaw, a Xiaomi account, and a Mijia camera.
This open attitude will quickly attract developers to improve the ecosystem, which in turn will further consolidate Xiaomi’s barriers.
The core confidence that Miloco 2.0 can achieve this is Xiaomi’s self-developed MiMo large model.
Xiaomi’s large-scale model capabilities have already quietly made it to the first echelon in the world.
In April this year, Xiaomi MiMo-V2.5-Pro ranked first in the comprehensive intelligence index of global open source large models on the list of the authoritative evaluation agency Artificial Analysis. The Agent special index ranked first in open source. Its comprehensive strength can compete head-on with top closed-source models such as Claude Opus 4.6 and GPT-5.4.
According to data from the OpenRouter platform on June 10, Xiaomi MiMo-V2.5’s single-week call volume reached 2.19 trillion, ranking second in the world, second only to DeepSeek.
Recently, Xiaomi announced that the UltraSpeed mode of MiMo-V2.5-Pro has achieved a trillion parameter model output speed exceeding 1,000 tokens/s, and a 10-fold speed increase only costs 3 times the price.
From large end-side models to AI-native mobile phones, from whole-house intelligence to car-machine AI, Xiaomi’s AI implementation has been very intensive in the past six months or so: in March, it launched the trillion-parameter end-side model MiMo-V2 Pro, in April it released HyperOS 2.0, and MiMo-V2.5 topped the open source list. In June, it open sourced MiMo Code and released Miloco 2.0.
Xiaomi plans to invest approximately 60 billion yuan in the AI field in the next three years, and this investment has begun to bear fruit in real products.
When AI enters the home, everyone’s biggest concern is always privacy issues.
If the cameras and sensors at home collect so much data, will it be leaked?
The answer Xiaomi gave this time is indeed stable enough.
Miloco 2.0 implements two layers of privacy protection: First, all original data is saved on the local side, disposable, and automatically cleaned up within 30 days. Users can clear it at any time with one click, and the original data will never be uploaded to the cloud. Second, the original data is completely isolated from the AI Agent. The Agent can only get the semantic results processed by Miloco, but not any original image, audio and video data.
Users can also set a blacklist for the camera and a whitelist for family members. The data is completely controlled by the user.
This approach strangles the bottom line of privacy and security and basically dispels users’ core concerns.
Xiaomi’s open source Miloco 2.0 this time is essentially a blueprint for whole-house intelligence in the Agent era.
In the past, everyone said that Agents would change productivity, but few people thought clearly about how Agents would enter ordinary people’s homes.
Xiaomi has given a path with practical actions: you do not need to buy a new house and replace all the equipment. Based on the existing Mijia ecosystem and an AI framework, you can turn a passive smart home into an active AI housekeeper.
More importantly, open source has lowered the threshold for the entire industry. Whether it is third-party hardware manufacturers or individual developers, they can develop new scenarios and applications based on Miloco.
In the future, more and more novel scenarios will be discovered. From health reminders to travel preparations, AI will thoroughly improve the service experience of the entire family.
In the past ten years, Xiaomi has defined the era of smart home interconnection. In the next ten years, Xiaomi wants to define the era of AI active services.
“You don’t have to take care of your family, your family will take care of you.” This sentence is no longer a line in science fiction movies. It will slowly enter every ordinary family in the next few years.
And those players who have built their AI brains and ecological barriers in advance will reap the biggest dividends from this wave of industry changes.
When 1.2 billion devices are activated by AI, the real GPT moment for smart homes may come sooner than we expected.
Are you ready to let AI help you manage your home?
