skulldc

RuView uses ESP32 WiFi CSI sensors to detect presence, breathing and pose without cameras

The MIT-licensed RuView project reads WiFi Channel State Information from low-cost ESP32 nodes to report presence, vital signs and activity, and its README separates real from unvalidated figures.

What it is

RuView, from the ruvnet/RuView repository, is a WiFi sensing platform written mainly in Rust and released under the MIT license. The README describes the idea: people who move, breathe or sit still disturb the radio waves a WiFi router already puts into a room, and RuView reads those disturbances as Channel State Information (CSI) from low-cost ESP32 sensors. From that signal it reports presence and occupancy, breathing rate and heart rate, activity such as walking or falls, room fingerprints, and sleep quality, without cameras or wearables.

The repository was created on 2025-06-07. In the GitHub trending snapshot it had 103357 stars, 245 of them gained in the last day, and 13397 forks; those figures are a point-in-time snapshot.

Hardware and models

According to the README, RuView runs on edge hardware: an ESP32 mesh, at as low as $9 per node, optionally paired with a Cognitum Seed for persistent memory, cryptographic attestation and AI integration, for a $140 total BOM. A pretrained model published on Hugging Face as ruvnet/wifi-densepose-pretrained fits in 8 KB when 4-bit quantized. The README says the v2 encoder reaches a held-out temporal-triplet accuracy of 82.3%, up from 66.4% raw, and that an older “100% presence” figure was measured on a single-class recording and has been retracted.

The project is open about which numbers are not yet product results. The committed on-device pose_v1 model is described as a first cut with PCK@20 = 3.0%, below the ADR-079 target, and its runtime path is still a stub. The 82.69% torso-PCK@20 figure belongs to a separate published MM-Fi benchmark model, ruvnet/wifi-densepose-mmfi-pose. The unified RF world model, which combines WiFi CSI, radar, UWB and cellular sensing, is labelled as still synthetic until real-data validation. Through-wall sensing is listed at up to ~5 m, signal-dependent.

Running it

The quick start offers a Docker image with simulated data, so no hardware is needed:

docker pull ruvnet/wifi-densepose:latest
export RUVIEW_API_TOKEN=$(openssl rand -hex 32) # required; the container exits 64 without it
docker run -p 127.0.0.1:3000:3000 -e RUVIEW_API_TOKEN ruvnet/wifi-densepose:latest

Other options flash ESP32-S3 or ESP32-C6 nodes for live sensing, add a Cognitum Seed, or install the Python package with pip install ruview. For smart homes, the README says RuView drops into Home Assistant with one --mqtt flag, or pairs into Apple Home, Google Home, Alexa or SmartThings as a Matter Bridge, and ships 21 entities per node.

Agent tooling and safety defaults

The repository also publishes a metaharness, @ruvnet/ruview, for humans and AI agents. It offers source-cited guidance, device access for CSI nodes, radar and LiDAR, firmware flashing with boot evidence, an MCP server over stdio or HTTP, a Claude Code mod, guarded Claude Code/Codex agents, deterministic verification and an honesty check for accuracy claims. The README lists its safety defaults: agent runs are read-only unless both --allow-write and --confirm are given, hardware reads need the device-access grant, flashing and calibration never run over the HTTP transport, and device-reported vitals are labelled as unvalidated.


Source: ruvnet/RuView on GitHub