An open-source platform for spatial awareness using WiFi signals: It reads Channel State Information (CSI) via ESP32, uses small pre-trained models to determine if people are in the room and what they are doing, and estimates breathing and heart rate trends. It can integrate with Home Assistant and Matter.
Stars 101072 (+557 / 24h)
Reasons
- Pre-trained models are published on Hugging Face at ruvnet/wifi-densepose-pretrained; the author claims that after 4-bit quantization, the size is only 8 KB. The author has retracted earlier claims of "100% presence detection" based on single-class recordings, now reporting 82.3% on an unlabeled holdout set.
- The @ruvnet/ruview toolkit provides an MCP server (stdio or HTTP), a Claude Code panel, and restricted Claude Code / Codex agents. Agents are read-only by default; writing files requires --allow-write and --confirm.
- Automatically integrates with Home Assistant via MQTT, or acts as a Matter bridge for Apple Home, Google Home, and Alexa.
Usage
- First, view the interface with simulated data: docker run -p 127.0.0.1:3000:3000 -e RUVIEW_API_TOKEN ruvnet/wifi-densepose:latest
- Real sensing requires CSI-capable hardware such as ESP32-S3; standard laptop WiFi can only perform presence detection based on signal strength. Use npx @ruvnet/[email protected] doctor to check your local environment.
Risks
- The author states: Vital signs reported by the device are unverified, accuracy figures for the unified RF world model are based on synthetic data, and the pure WiFi 17-point pose model is still in its initial version. These are research prototypes, not medical devices.
- The README contains many promotional claims (wall-penetration, search and rescue, prison monitoring, etc.); capability boundaries should be assessed by reviewing the documentation for each module. Sensing the presence and vital signs of others through walls raises privacy concerns, and consent must be obtained before deployment.
- Full functionality is recommended with Cognitum Seed hardware (the author claims approximately 140 USD per set). The repository has existed since 2025, and the license is MIT.