The platform is an operating system.
A local intelligence kernel. It senses, remembers, reasons -- leaves an auditable trace at every step, and stays open for third parties to extend.
StarWeave senses · MoonGuide remembers · YaoHeng decides
The core of AIOS is not a single model. It is three cooperating engines — StarWeave for perception fusion, MoonGuide for the memory stack, YaoHeng for confidence-based decisions — together supporting how the space perceives, remembers, and judges.
StarWeave
Processes vision, voice, sensors and spatial events, fusing multimodal signals into one coherent perception of the space.
MoonGuide
A four-layer stack across L0 static, L1 live, L2 habit, L3 semantic, moving the system from reactive replies to understanding habits and proactive suggestions.
YaoHeng
Fuses multi-source evidence with a confidence score, used as the basis for downstream action — never a single model executing on its own.
Sense · Remember · Reason · Decide · Federate.
Sensing
Multiple channels in parallel -- vision, sound, temperature, light, and sensor events. It reads the whole room, not a single video stream. Whether vision runs, and what is stored, follow on-site policy and audit settings; answers about state are grounded in devices and sensors.
- · Works with mainstream ecosystems, no forced rip-and-replace
- · Critical paths stay local, with optional cloud toggles
- · Query answers are grounded in real sources, not invented figures
Memory
From the moment to the long-term, it builds up real memory -- who is in which room, whose rhythm shifted this week, which lighting you have favored lately.
- · Layered memory, from now to habits
- · Not only what happened, but why
- · Memory never leaves your home
Reasoning
A local brain, thinking inside your home. Every action gets rehearsed before it runs, and leaves behind a trace anyone with permission can replay.
- · High-risk actions need approval
- · Shadow runs before live
- · Structured traces for replay and export under policy
Decision
Reasoning lists what could happen; a decision lands on what does. The local YaoHeng engine scores each candidate by evidence, risk, and the rules of your house -- and keeps the why next to the choice.
- · Every action ships with confidence and risk
- · When unsure, it pauses and asks
- · Both the choice and the reason are replayable
Federation
Industry and site profiles move as offline, verifiable bundles between sites; imports preview first, then an admin explicitly applies them to the fact base and audit log. Identities and sensitive detail stay behind.
- · Offline bundles and checksums for air-gapped or controlled copy
- · Align rules and profiles across sites without resident-level payloads
- · RBAC-style views -- families see what policy allows
MoonGuide · Memory that remembers you
From the immediate moment to years of habit, one layer at a time. It is not storage. It is understanding, slowly, who you are and what you mean.
L0 · Static facts
Rooms, device topology, ownership, and labels -- slow to change, yet the anchor for every query and control path.
L1 · Live state
Who is where, what is on, temperature, humidity, and light -- refreshed moment by moment as events arrive.
L2 · Habits
When that lamp usually goes off, which lighting feels like weekend -- rhythms learned from repeated behavior.
L3 · Semantics & intent
Goals and context in language -- vectors and tags that align who you are and what you mean with retrievable memory.
Every step, traceable
Every thought, every action, every rule change leaves a line, like a ledger. Family members see last night. Managers replay last week. Regulators export exactly what they are owed -- no more, no less.