Gaia turns context into the operating system

Xiaomi's XiaoAi Memory turns a three-finger swipe on any screen into a saved piece of personal context, while Hisense's JUOS aims to extend that idea across an entire home. ifanr reports that saved material can be retrieved by describing it to XiaoAi. It also reports that memories and conversations from a Xiaomi phone can sync to the Xiaomi Interconnect app on a PC, along with notes and photos. The operating system is becoming a layer that remembers what matters across screens.

A Chinese outlet describes JUOS, released on August 31, as a family-oriented AIOS that uses Super Xiaojü to coordinate connected home devices. That shift raises the harder question: when an agent acts in a shared space, whose memory should guide it? Whose preferences and authority should guide it?

A personal context layer across screens

Xiaomi's personal-context model begins with material already on a screen. XiaoAi Memory lets a user swipe up with three fingers on any screen, prompting the system to recognize and save what is displayed to Super XiaoAi's memory, according to ifanr. That saved item can later be retrieved by describing it to XiaoAi. The point is continuity: a page or fragment does not need to remain open to remain available to the assistant.

The assistant is meant to travel with the user's work rather than stay on one device.

ifanr reports that memories and conversations from a Xiaomi phone can sync to the Xiaomi Interconnect app on a PC. Notes and photos can sync as well. Super XiaoAi is also available on Windows and Mac computers. It can run scheduled tasks. These include checking email and refreshing calendars at fixed times each day. It can also interpret smart-home requests and break them into requirements. It can then execute the resulting tasks. Together, those functions cast the system as a personal layer spanning screens and devices.

Hisense's JUOS points at a broader version of the same idea. That version is more difficult. Released on August 31, JUOS is described by the outlet as a family intelligent-companion-level AIOS. It is intended to connect smart devices across home scenarios, rather than serve only one device. Its Super Xiaojü system coordinates devices throughout a home. Xiaomi's model can begin from one person's saved material and conversations. A household system must operate across a shared space, where the relevant context cannot automatically be assumed to belong to one person.

Shared memory is harder than synced data

Syncing information between screens is not the same as giving every app access to one usable memory. According to ifanr, memories from a Xiaomi phone can sync to the Xiaomi Interconnect app on a PC. Conversations can sync as well. So can notes and photos. That creates continuity between devices. Yet the recording tools show where continuity can stop: the Notes app records audio, identifies speakers, then produces documents and meeting minutes; Recorder provides live subtitles and AI summaries.

Both use MiMo-V2.5, but ifanr says their data layers are not fully connected.

That separation matters because a model can understand material without being able to carry it, with the right permissions, into another task. A transcript in Recorder and a meeting record in Notes may be generated by the same foundation model, but they are not automatically one shared context. The promise of cross-device memory therefore depends on more than synchronization. It depends on which application owns the underlying data and which system is allowed to retrieve it.

Qixidi frames the home version as a deeper integration problem. One source describes its Gaia architecture as one spatial system that brings switches together from sensing through execution. It also covers lighting and air. Energy, sensors, and agents are part of the same system. A home computing hub performs local computing. It provides storage and supports system operation, while a space controller connects different power-supply methods and communication protocols. Qixidi also says its HClaw agent, Xiaoqi, can produce energy-consumption analyses after electricity-use data from key devices is collected centrally. Here, shared memory is not merely content appearing on another screen; it is operational data that can inform a physical system's next action.

A household needs identity and authority

Xiaomi's XiaoAi Memory is built around a person deliberately capturing context. According to ifanr, a user swipes up with three fingers on any screen, prompting the system to recognize and save what is displayed to Super XiaoAi's memory. The user can later retrieve that material by describing it to XiaoAi. This makes context portable for its saver, yet the description does not say what happens when somebody else can see or hear the retrieval.

A shared room makes identity part of the operating system.

According to a source, Qixidi's Xiaoqi uses sound-source localization, human-body sensing, voiceprint recognition, and spatial-state information before adjusting devices. It is meant to identify both the speaker and that person's location. JUOS takes another route: its personalized desktop can identify the person before the screen through facial or voiceprint recognition, then offer older users opera and news, or give children age-appropriate content with viewing reminders. Younger users can receive resumed series playback and pre-game reminders.

Recognition, however, is not authority. None of these descriptions explains how a system resolves conflicting preferences, or what it does after a mistaken identification. The woshipm author argues that memory ownership and the question of speaking information aloud around others are separate risks. The author recommends classifying memories by sensitivity. Relationship material can be shifted toward generalized wording, a private screen, or delayed disclosure when another person is detected. The same approach can apply to name material, health material, and finance material.

Invisible automation still depends on sensing

Qixidi's version of an AI-native residence is meant to be adaptable, able to grow, and unobtrusive, according to a report. Its AI Super Constant Climate System continuously senses conditions in a space, then coordinates systems that bring in fresh air. It also manages exhaust air and temperature control. Shen Yanan describes the result as a comfort bubble that follows people.

That promise depends on sensing people as well as rooms.

JUOS illustrates the trade-off. A report says its AI personalized desktop can identify the person before a screen through facial or voiceprint recognition, while JUOS also claims to detect tone and emotion for a more personable response. Recognition can make a shared home feel responsive. It can also make household members and children visible to a system they did not actively address. Visitors may become visible to it as well. Cleaners may too. So may guests.

Consent therefore cannot be reduced to a setting that permits recognition. The woshipm author argues that memory ownership and the question of whether information may be spoken aloud around other people are distinct risks on shared devices. A system may know whose memory it holds. It may still expose it at the wrong moment through a voice response.

The proposed safeguard is disclosure-aware memory. woshipm recommends sensitivity classes, with relationship content treated as high sensitivity by default. Name content receives the same treatment. Health and finance content are also high sensitivity by default. If another person is detected, the agent should use generalized wording. It can move the result to a private screen or delay disclosure. That makes unobtrusive operation mean more than quiet automation: it means knowing when silence or restraint is the useful action.

Even command handling can limit unnecessary exposure. woshipm recommends three processing layers, including direct execution for clear requests such as playing a Jay Chou song, without sending that request to a model. The less a household interaction needs to become retained interpretation, the narrower its consent burden can be.

Choosing an AIOS workflow: personal context versus shared-space control

• You want a personal assistant to capture material from a phone and retrieve it later on a computer. Use Xiaomi's personal-context workflow: XiaoAi Memory can save on-screen content with a three-finger swipe, retrieve it through a description to XiaoAi, and sync memories, conversations, notes, and photos from a Xiaomi phone to Xiaomi Interconnect on a PC. Super XiaoAi is available on Windows and Mac computers.
• You need repeatable personal productivity tasks rather than a one-off chat response. Use Super XiaoAi for scheduled tasks such as checking email and refreshing calendars at fixed times each day. Reserve Expert Mode for work that warrants its point cost: users receive 1,000 free points each month, while Xiaomi estimates 12-20 points per conversational task and 100-300 points for complex light-office tasks such as PDF and PPT generation.
• The request is a clear, low-ambiguity command in a shared device environment. Prefer direct execution over model mediation where possible. One proposed design approach recommends directly executing clear commands such as "Play a Jay Chou song" without sending them to a model, reducing unnecessary interpretation for routine actions.
• An agent will change devices or conditions in a home used by multiple people. Choose systems that establish who is present before acting, and treat authority and disclosure as separate problems. Qixidi says its Xiaoqi agent uses sound-source localization, human-body sensing, voiceprint recognition, and spatial-state information to identify who is speaking and where they are before adjusting devices. For sensitive memories involving relationships, names, health, or finances, the recommended default is generalized wording, a private screen, or delayed disclosure when another person is detected.
• You want an AI layer to coordinate a whole home rather than improve a single screen. Consider a home-oriented AIOS only when the required devices are already in its ecosystem. Hisense says JUOS links televisions, projectors, and smart home appliances, and can use a living-room screen to control connected household devices by voice after linking with the Hisense Smart Home ecosystem. Qixidi's Gaia is designed around switches, lighting, air, energy, sensors, and agents as one spatial system; its home computing hub is described as handling local computing, storage, and system operation.

Test the controls behind the AIOS promise

A practical AIOS test begins with a task already in motion. On Xiaomi, XiaoAi Memory can save displayed content from any screen with a three-finger swipe and retrieve it from a description. According to ifanr, it can sync memories from a phone to the Xiaomi Interconnect app on a PC. It also syncs conversations. It syncs notes. It syncs photos. Check whether the recovered item retains its meaning, not merely its text.

Then test the handoff in a shared room.

A cited report says JUOS can continue partially watched small-screen content on a living-room large screen. It can also recognize NAS devices. It can recognize set-top boxes. It can recognize game consoles, matching picture quality and controls. Those are useful context-transfer checks: move between the devices actually used, then inspect whether the system selects the right source. Inspect whether it selects the right control scheme. Inspect whether it preserves state without asking for repeated setup.

Memory needs an equally concrete audit. The woshipm author recommends keeping unconfirmed memories out of later conversations and hiding failed writes from the remembered record. Ask what can be corrected. Ask what can be deleted. Also ask what can be prevented from entering memory. On shared devices, ownership is one risk. Permission to speak information aloud is another, woshipm argues. Sensitive material involving relationships may need a private screen. It may instead need delayed disclosure. The same may be true of names. It may be true of health. It may be true of finances.

Withhold broad household authority. Wait until per-person controls are usable. Super XiaoAi can schedule email checks and calendar refreshes, while its smart-home control can interpret a request and execute tasks, according to ifanr. Yet the available descriptions do not specify cloud retention, outage behavior, interoperability exit paths, or action-reversal tools.

Three AIOS approaches to context, identity, and action

Start with a handoff test: record a meeting in Notes, then check whether its speaker-labelled transcript and document can be used where Recorder's live subtitles and automatic summary are needed. Check the minutes separately. According to ifanr, both apps use MiMo-V2.5, yet their data layers are not fully connected. Ask where context stops, and require a way to inspect and correct what crossed the boundary.

Watch the meter, too. ifanr says Expert Mode uses 12-20 points per conversational task, while PDF and PPT generation uses 100-300 points each.

In shared work, keep automation authority narrow until guest access and per-person preferences have distinct controls.

For readers outside China

• Availability: The source material does not say whether Xiaomi HyperOS 4, Super XiaoAi, Hisense JUOS, or Qixidi Gaia are available outside China. Super XiaoAi is stated to be available on Windows and Mac computers. Hisense says JUOS will be pushed to a first batch of supported models starting in September and then to subsequent models in batches; the upgrade covers recent Hisense U, E, and A series televisions, laser TVs, Vidda televisions, and Vidda projectors.
• Pricing: Xiaomi offers a paid Super XiaoAi subscription. Entry costs 29 yuan per month, or 19 yuan per month with continuous monthly subscription, for 4,000 points per month. Standard costs 59 yuan per month, or 39 yuan per month with continuous monthly subscription, for 10,000 points per month. Enhanced costs 149 yuan per month, or 99 yuan per month with continuous monthly subscription, for 34,000 points per month. Users receive 1,000 free points each month. Pricing for JUOS, Gaia, and Qixidi's ANDI is not disclosed in sources.
• Closest Western equivalents: No direct Western equivalent for these cross-device and whole-home AIOS systems is identified in the source material.; Replika and Character.AI are the named Western AI-companion reference points, but they are not presented as equivalents to Xiaomi, Hisense, or Qixidi's device-and-home orchestration systems.
• Data residency: Qixidi says Gaia includes a home computing hub for local computing, storage, and system operation. Beyond that architectural claim, the source material does not cover where personal data is stored, whether it leaves the home, cross-border transfers, retention periods, or user controls over voiceprints, facial recognition, and memory data.

Sources

• woshipm 当AI成为最听话的那个人，我们究竟在监管什么？ https://woshipm.com/ai/6441649.html
• 36kr 全球首个AI原生智能住宅：成为智慧生命体的家是什么样？ https://36kr.com/p/3974490866561280
• ifanr 首发 | 澎湃OS 4 体验：小米第一个 AIOS，用起来怎么样？ https://ifanr.com/1675208
• woshipm 从 0 设计 AI 陪伴产品：陪伴产品的内核是什么？ https://woshipm.com/pd/6457521.html
• 36kr 科氪 | 海信 JUOS，让家里每块屏从此巨简单、更懂家 https://36kr.com/p/3967193568500994
• woshipm AI颠覆社交是一个伪命题 https://woshipm.com/ai/6457604.html