Fullive.ai Somni: The AI Bedside Sleep Device Built to Learn Night by Night

Somni is Fullive.ai’s first sleep product: a screen-free, contactless bedside system designed to coordinate sound, light and scent through the night rather than simply record sleep metrics. The central idea is response learning—using repeated observations of a person’s reactions to refine later environmental interventions. 410

What Somni is designed to do

Fullive describes Somni as a bedside system that works without a wearable or a display. Reported hardware includes millimeter-wave radar, a microphone array and environmental sensors. Together, these are intended to monitor signals such as breathing, heart rate, movement and bedroom conditions. 47

The device’s interventions are deliberately environmental rather than invasive:

• Sound for settling and overnight support
• Light for wind-down, nighttime adjustments and a gentler wake-up
• Scent coordinated with the rest of the sleep environment 410

The company says Somni aims to help users settle, remain asleep and wake more naturally. Those are intended product outcomes, however, not evidence that it treats insomnia or another medical condition. 10

How “response learning” works

Most sleep trackers focus on describing what happened overnight. Fullive’s stated goal is different: to learn which controllable changes appear to help a particular person in a particular state.

The proposed loop is:

1. Observe state: collect physiological and environmental context during sleep.
2. Apply a bounded intervention: adjust a modality such as sound or lighting.
3. Measure response: assess subsequent physiological and behavioral signals.
4. Refine the next plan: retain or alter the intervention on later nights based on the observed response. 56

Fullive characterizes each night as a small experiment. If one sound-and-light combination does not appear to produce the desired response, the next night’s plan can change; if it appears helpful, that feedback becomes evidence for future selection. 5

In the company’s terminology, the useful dataset is “state → intervention → response.” The aspiration is to learn more than a user’s baseline sleep patterns: it is to estimate how that person may respond when the sleep environment changes. 56

The “response model” and Bio-OS ambition

Fullive frames Somni as an early node in a broader AI4Bio strategy that combines models with hardware deployed in real-world spaces. The company’s reported architecture is “sensors × foundation model × Bio-OS Harness.” Sensors collect bodily and room signals; a model proposes individualized strategies using physiological context; and the Bio-OS layer coordinates sensing and physical outputs in a closed loop. 6

That leads to two related concepts:

• A response model, intended to predict how an individual may react to a bounded intervention in a given physiological state.
• A longer-term physiological world model, intended to represent how physiological states change over time under context and interventions. 6

The latter is an ambitious research direction, not a demonstrated Somni capability. In biomedical AI research, world models generally refer to models that represent states and their evolution under perturbations, treatments or environmental changes. 1

Funding and team: what has been reported

Fullive.ai was founded in Beijing in 2026 and describes itself as an AI4Bio company pursuing a “model × body” approach—pairing AI health models with deployable hardware. 6

Reporting says the startup completed three funding rounds in less than a year, with Hillhouse Capital participating in all three. A later round was reported as led by China Merchants Venture Capital, with other investors participating. The company said funding would support Somni research and development, production progress and longer-term AI4Bio work. 116

Public reporting also describes a team spanning AI, hardware and commercialization experience, although detailed staffing claims should be treated as company- and media-reported background rather than independently audited evidence.

Safety, privacy and the evidence gap

A product that listens to and acts within a bedroom needs clear boundaries. Fullive says Somni’s interventions are constrained to low-disturbance environmental modalities—sound, light and scent—and that model actions operate within predefined limits. 6

The company also describes a device–cloud approach in which raw radar waveforms are preferentially kept locally, while cloud-bound information is encrypted and minimized to reduce privacy exposure. 6

Just as important is what is not established by the available reporting. Somni had reportedly completed small-batch trials and was moving toward production and larger-scale clinical validation in 2026. 46 The provided sources do not establish that it has completed large clinical trials, received medical-device authorization or demonstrated treatment efficacy for insomnia.

Why sleep is the starting point

Sleep offers Fullive a comparatively repeatable environment: long periods of passive observation, recurring nightly routines and a constrained set of environmental actions. That makes it a practical first setting for testing a sense–intervene–measure loop.

Fullive’s larger roadmap is to extend that loop beyond sleep into recovery, stress and daytime settings. 69 For now, those broader applications remain a company ambition. Somni is the nearer-term test: whether a contactless bedside device can safely personalize a sleep environment using real-world feedback over time.