WiFi vision (WiFi sensing) uses a WiFi router to see through walls: it sends wireless signals through walls, reads how they change when they bounce off people, and uses AI to rebuild a person’s physical shape plus breathing rate and heart rate. The practical target is passive monitoring, and the clearest use case is an elder care home that watches vitals 24-7 and alerts a real nurse when something goes wrong.
Kevin walks through it in a KevBuildsApps video: WiFi Vision: See Through Walls With Claude in 2026. The project it references is RuView, an open-source project by ruvnet, not one of Kevin’s own repos.
WiFi router
|
| radio waves pass through the wall
v
[ wall ] ---> person in the next room
^ |
| | waves come back changed
| v
signal changes --> AI model
|
+------------+------------+
| |
v v
body pose / shape breathing + heart rate
| |
+------------+------------+
|
v
threshold or anomaly
|
v
alert a real nurseWhat Is WiFi Vision and How Does It See Through Walls?
WiFi vision is the practice of reading changes in ordinary wireless signals to infer what is happening in a room you cannot see, including through a wall.
The mechanism is simple to describe and hard to do well. A router transmits radio waves. Those waves travel through drywall and timber, hit a person, and reflect back. A human body is mostly water, and water absorbs and shifts radio energy, so the returning signal carries the imprint of the body that reflected it. Movement changes that imprint every fraction of a second.
The AI part is the pattern reader. Instead of treating the signal as noise, the model learns the relationship between specific shapes of signal change and specific body positions, then reconstructs a pose. The same class of model picks out the much smaller periodic changes caused by a chest rising and falling, or a heartbeat nudging the signal.
Two properties matter for anything real:
- It is passive. Nobody wears a device or holds still for a camera. The sensing happens because the router is already broadcasting.
- It does not need line of sight. A camera needs a clear view. A router does not, which is the entire reason elder care is the interesting use case.
How Claude Plus a WiFi Router Turns Signals Into Pose Tracking
The router produces raw signal measurements. Claude is the layer that turns those measurements into something a person can act on.
A working loop looks like this:
- The router collects channel measurements, essentially snapshots of how the signal is behaving across frequencies.
- A model converts each snapshot into a pose estimate, a skeleton or body shape.
- Claude receives the structured output, not the raw waveforms: a timestamped record of position, posture and movement.
- Claude reasons over that record in plain language and decides whether the situation is normal or worth flagging.
That split is the useful engineering decision. Signal processing stays in code, where it belongs. Interpretation moves to the model, which is good at context. A prompt in that loop might look like:
Here are the last 15 minutes of pose readings for room 4.
The resident was standing until 02:14, then the body shape
drops to floor level and stays there. Breathing signal is
present but shallow. Decide if this is an emergency and
write the alert a night nurse should read.That is the difference between a sensor that beeps and a system that tells a nurse what happened. If you want the same shape of workflow applied to your own accounts, the walkthrough in how to build a social media agent with Claude Code shows the same pattern: raw data in, model interpretation, an action at the end.
How WiFi Sensing Reads Breathing Rate and Heart Rate
Breathing and heart rate fall out of the same signal as a much smaller, much more regular wobble, and reading them is a filtering problem before it is an AI problem.
A chest moving up and down shifts the reflected signal on a cycle of roughly 12 to 20 times a minute at rest. A heartbeat moves the body surface far less but still moves it, several times a second. Those two patterns sit at different frequencies, so a model can separate them from the larger body movement signal, then estimate a rate.
| Layer | What it reads | Use it for | Main weakness |
|---|---|---|---|
| Gross signal change | Presence, motion, falls | Did someone fall, did someone leave | Confused by pets and moving furniture |
| Pose estimate | Body shape and posture | Standing, sitting, lying, gait | Needs training data for the room |
| Fine periodic signal | Breathing rate | Resting monitoring, sleep | Drops out when the person moves |
| Micro signal | Heart rate | Trends over time | Noisy at distance, needs stillness |
The honest reading of that table: presence and pose are solid, breathing rate is workable, heart rate is the weakest link. Treat heart rate as a trend signal that prompts a check, not as a replacement for a medical device. Kevin’s video is clear that the goal is alerting a real nurse, not diagnosing anyone.
Can WiFi Vision Really See Through Walls? What It Can and Cannot Do
WiFi vision does pass through walls, but the answer to “can it see through walls” is yes with limits, and the limits decide whether a deployment works.
What works in its favour:
- Radio waves at WiFi frequencies pass through drywall, wood and glass with modest loss.
- No light is needed, so it works in the dark and behind a closed door.
- Nothing has to be worn, charged or remembered.
What gets in the way:
- Dense materials. Concrete, brick, metal lath and foil-backed insulation attenuate the signal heavily. A room behind a load-bearing concrete wall is a different problem from a room behind drywall.
- Other movement. Ceiling fans, curtains in a draught, pets and even a running shower add signal change that is not a person.
- Several people in one room. Separating two bodies from one signal is much harder than tracking one.
- Resolution. This is not a photograph. It reconstructs a coarse body shape and pose, not a face or a readable screen.
The practical conclusion: one router per room, calibrated to that room, with the model retrained or tuned on site. That is exactly what the elder care deployment in the video describes, a router in every room rather than one router hoping to cover a corridor.
The Passive Elder Care Use Case: A 24-7 Virtual Nurse
The reason this technology gets attention is that it solves a staffing problem no camera can solve politely.
Picture a care home at 3am. A resident gets out of bed, walks to the bathroom, and falls. Nobody is awake to see it. With a router in the room, the sequence is visible in the signal: standing, walking, then a body shape at floor level that does not get up. Breathing is still detectable, so the system knows the person is alive. An alert reaches the night nurse with the room number and what happened.
Why this beats the alternatives in a care setting:
- No camera in a bedroom. A resident is not filmed undressing or using the toilet. That single fact decides most family conversations.
- No wearable to forget. Residents with dementia take devices off. A router stays on the wall.
- It runs continuously. Vitals get watched all night, not only during a scheduled check.
Kevin framed the goal plainly in the video: when something goes wrong, it alerts a real nurse to respond fast. The AI is the watch, the nurse is the responder. Watch the demo on the KevBuildsApps launch video if you want to see how Kevin builds and explains projects like this.
How to Set Up the Claude Plus WiFi Project Step by Step
The setup is a router, a collector, a model and Claude, wired in that order, with each stage testable on its own.
The source material is the RuView repo at github.com/ruvnet/RuView. Read its own README for install steps, dependencies and hardware notes. Nothing below replaces that file.
- Get one router you can read signal data from. Not every consumer router exposes what you need. Start with the hardware the RuView README recommends.
- Collect a baseline. Record the signal in the room with nobody in it. This is your reference, and every later reading is compared against it.
- Record labelled movement. Capture a few minutes of one person walking, sitting and lying down, so the model has examples of what each pose looks like in this specific room.
- Run pose estimation and check it visually. Before adding anything clever, confirm the reconstructed skeleton tracks the person. If it does not, nothing downstream will work.
- Add the breathing and heart rate layer. Filter for the periodic components and plot the rates over time. Expect dropouts when the person moves.
- Put Claude at the end, not the start. Feed it structured readings plus a clear decision rule, and have it write the alert. Do not ask it to interpret raw waveforms.
- Test the failure path first. Simulate a fall and confirm the alert reaches a human. A monitoring system that has never fired an alert is untested.
If you are new to wiring Claude into an agent loop, building a content repurposing agent with Claude Code covers the same plumbing from the other direction: data in, model reasoning, a scheduled action out. The KevBuildsApps channel at youtube.com/@KevBuildsApps has the build videos.
WiFi Vision vs Cameras and Wearables for Health Monitoring
Each option wins on a different axis, so the right answer depends on whether you are optimising for accuracy, privacy or whether anyone will actually keep wearing the thing.
| Approach | Strengths | Weaknesses | Good fit |
|---|---|---|---|
| WiFi sensing | No line of sight, no wearable, works in the dark | Needs a router per room, sensitive to dense walls and extra movement, coarse resolution | Overnight presence, falls, breathing trends |
| Cameras | High detail, easy to verify an event | Constant filming, uncomfortable in bedrooms and bathrooms | Common areas, entry points |
| Wearables | Direct contact measurements | Must be worn and charged, easy to remove | Residents who keep them on |
| Bed and floor sensors | Reliable, cheap, localised | Only covers the spot it is installed in | Bed exits, bathroom trips |
The pattern most care homes will land on is layered: WiFi sensing for continuous room coverage, plus one or two targeted sensors where absolute certainty matters. WiFi sensing is not trying to replace everything. It fills the gap that cameras and wearables both leave open.
FAQ
Can WiFi really see through walls?
Yes, within limits. WiFi radio waves pass through drywall, wood and glass, and a model can read the changes in those waves to infer a person’s presence and body shape on the other side. Concrete, brick, metal and foil-backed insulation weaken the signal enough that a room behind them needs its own router and its own calibration.
Is WiFi sensing accurate enough for heart rate and breathing monitoring?
Breathing rate is the more reliable of the two, because a chest movement changes the signal on a clear cycle. Heart rate is measurable but noisier, especially at distance or when the person moves, so it is best treated as a trend that prompts a check. Neither should be presented as a substitute for a medical device.
What is Claude reference and how does it fit into WiFi vision projects?
“Claude reference” is search shorthand for what people mean when they ask how Claude is used in this kind of project, and the answer is as the reasoning layer. The router and the signal model produce structured readings, then Claude interprets those readings and writes the alert a nurse reads. It does not process raw radio waveforms, and asking it to would be the wrong design.
Do I need Claude skills to build a WiFi sensing setup?
No. Claude skills are reusable instruction packages that help an agent repeat a workflow consistently, and they are useful once you want the same analysis run the same way every time. For a first build, a clear prompt and a fixed output format get you most of the way.
Is WiFi vision monitoring a privacy risk in elder care homes?
It is less invasive than a camera in the same room, because no image is produced and the output is a body shape plus vitals, not a picture of a person. That does not remove the obligation to get informed consent from residents and families, and to say plainly what is collected, who sees alerts and how long readings are kept.
Get started
Kevin’s WiFi vision video and the RuView repo are the places to start if you want to build the sensing side yourself. If you would rather put an AI agent to work on something running today, Creator OS ships a hosted MCP server and a CLI so Claude can run your accounts, and it covers 11 platforms plus Skool in one workspace.
Creator OS plans start at $14.99/month for 11 platforms plus Skool, with the API, the MCP server and the CLI included. Start at https://www.creatoros.ca/sign-up, and read the docs at https://www.creatoros.ca/docs for the setup details.
Get started for $0
Free trial for new subscribers. Cancel anytime.