RuView is an open-source platform that turns ordinary WiFi signals into contactless sensing — detecting whether a room is occupied, estimating body pose and vital signs, and recognizing activity, all without a camera or a wearable. Written primarily in Rust and released under the MIT license, it has drawn more than 83,000 GitHub stars and topped GitHub's trending charts by pushing a genuinely unusual idea into practical, low-cost hardware: the radio waves already bouncing around every home carry enough information to sense people, and cheap microcontrollers plus machine-learning models can decode it. RuView packages that pipeline into something a developer can actually run. ## What It Is WiFi routers constantly emit radio waves that scatter off human bodies. RuView uses inexpensive ESP32 sensor boards to capture Channel State Information (CSI) — fine-grained phase and amplitude readings across 56 subcarriers per channel — and feeds that raw signal through a processing and inference stack that produces spatial intelligence. The goal is sensing that works through walls and in complete darkness, with no video and no device worn by the person being sensed. Because it reads the WiFi that already exists in a space, the sensing is passive and privacy-preserving in the sense that no images are ever captured. ## From Radio Signal to Spatial Intelligence The pipeline fuses multiple bands — three channels of 56 subcarriers each, giving 168 virtual measurements — then applies classical signal processing such as bandpass filtering, phase unwrapping, and spectral analysis to isolate the tiny periodic motions of breathing and heartbeat. On top of that sits the project's RuVector backbone, which combines attention networks, graph algorithms, and contrastive embeddings, and a pretrained encoder that emits 128-dimensional embeddings. That model ships as a compact 8 KB quantized artifact on Hugging Face, and the maintainers report 82.3 percent held-out accuracy on presence detection and 82.69 percent on 17-keypoint pose estimation against the public MM-Fi benchmark. ## What It Can Sense RuView targets a broad set of sensing tasks from the same signal source. It detects presence and occupancy through walls in real time, estimates breathing and heart rate within stated ranges, and performs 17-keypoint pose estimation. It advertises sub-200-millisecond fall detection using phase-acceleration thresholds, multi-person counting with adaptive normalization, sleep-stage classification and apnea screening, environmental fingerprinting for room identification and anomaly detection, and activity recognition such as walking, sitting, and gestures from temporal patterns. The intended applications span elderly-care fall detection and vital-sign monitoring, retail occupancy and dwell-time analytics, building HVAC and utilization automation, and industrial worker-safety zones. ## Hardware and Integration A defining feature is how cheap the entry point is. A minimal setup uses an ESP32-S3 board costing roughly nine dollars, with two to four nodes recommended for room-level coverage; an ESP32-C6 with WiFi 6 is an alternative, and research-grade CSI can be captured with Intel 5300 or Atheros NICs for full 3x3 MIMO. Firmware is flashed with esptool, after which nodes stream CSI to the platform, and a Docker image lets you try the stack on simulated data before buying hardware. Integration is a strong point: RuView works natively with Home Assistant via MQTT auto-discovery and bridges to Apple, Google, and Alexa ecosystems through Matter, and a Python client is available via pip. ## Usability in Practice For makers and developers, the appeal is a full, documented path from a nine-dollar board to a working sensing feed, plus a simulated-data Docker image to evaluate the software with no purchase. The maintainers back the project with unusual engineering hygiene for its category — the README cites over a thousand commits, hundreds of releases, 182 architecture-decision records, and deterministic proof scripts meant to make the core signal-processing claims reproducible. The honest caveats are significant and, to their credit, disclosed. RuView is explicitly beta software with changing APIs; a single ESP32 has limited spatial resolution so multiple nodes are advised; camera-free pose accuracy is still low (about 2.5 percent PCK@20) with camera-supervised fine-tuning in progress; through-wall range is roughly five meters and signal-dependent; older ESP32 and ESP32-C3 boards are unsupported; and an earlier "100 percent presence" figure was retracted as having been measured on a single-class recording. ## Pros and Cons The strengths are a genuinely novel, camera-free sensing approach; very low hardware cost; a broad range of sensing tasks from one signal source; native Home Assistant and Matter integration; a simulated-data Docker path for evaluation; and notably transparent documentation, benchmarks, and retractions. The trade-offs are that it is beta and fast-moving, that accuracy figures are mid-range and pose estimation remains weak without camera supervision, that reliable results need multiple CSI-capable nodes rather than a laptop's coarse RSSI, and that through-wall range and performance depend heavily on the environment. ## Outlook WiFi sensing has been an active research area for years, and RuView's contribution is less a scientific breakthrough than a serious attempt to make it deployable on commodity hardware with mainstream smart-home integration. The candor of its benchmark reporting, including a public retraction, is a good sign for a field prone to overclaimed demos. Whether it becomes broadly dependable will hinge on accuracy gains — especially in pose — and on how well the beta stabilizes into something non-experts can trust for safety-critical uses like fall detection. ## Conclusion RuView is an ambitious, refreshingly transparent take on turning WiFi into a contactless sensing layer, and its low-cost hardware and smart-home integrations make it approachable for developers and makers who want to experiment. It is best understood as capable, well-documented beta software rather than a finished product: promising for occupancy, activity, and vital-sign sensing, but one to evaluate carefully — with realistic expectations and multiple nodes — before relying on it for anything safety-critical.