“"Droll that the first 'Feature' listed is that it's 'Privacy First'." — treetalker, Hacker News (https://news.ycombinator.com/item?id=47302894, 2026-03-08)”
You want to monitor a room for falls, track occupancy in a building, or detect breathing anomalies during sleep — but cameras violate privacy, wearables need charging, and pressure mats do not scale. WiFi signals already pass through every room in a building, and the physics of signal disturbance by human bodies is real, peer-reviewed science. The gap is a production-ready stack that turns commodity WiFi hardware into a passive, camera-free human sensor without requiring body contact or line of sight.
WiFi routers continuously send signals that bounce off walls, furniture, and human bodies. An ESP32-S3 microcontroller running custom promiscuous firmware captures those signal disturbances as Channel State Information — 90 per-subcarrier amplitude and phase readings per packet, far richer than the single RSSI value consumer software exposes. A Rust DSP pipeline applies phase correction, outlier rejection, and Fresnel zone modeling to clean the signal. The cleaned data feeds into WiFlow, a 1.8M-parameter neural network (TCN + axial attention) that maps the signal pattern onto 17 COCO body keypoints, breathing rate, or heart rate. A Docker container on port 3000 serves the results as a web interface.
If you are an embedded systems or Rust engineer researching WiFi CSI and RF-based sensing, the DSP pipeline and WiFlow architecture are worth reading as a codebase reference. If you need a production fall detection or occupancy system today, this is not ready: open firmware crash bugs exist in the CSI capture path (#396, #438), zero independent validation has surfaced, and the hardware requirement contradicts the '$8' headline. Not useful for consumer WiFi adapters or standard laptops — those expose RSSI only, not CSI.
Worth reading the codebase if you are researching RF-based sensing — the WiFi CSI pipeline and WiFlow architecture reflect real academic techniques. Not worth deploying: the firmware has active crash bugs in the CSI capture path (#396, #438), the 92.9% PCK@20 figure was measured with camera-supervised training rather than a fully camera-free setup, and no external party has independently demonstrated the full stack on physical hardware as of 2026-04-29.
Deep-dive insight, Easy and Pro modes, plus action playbooks — the full breakdown is one tap away.