What it does RuView reads Channel State Information (CSI) from cheap ESP32 sensors to detect people through walls, count them, estimate 17-keypoint poses, and even extract breathing rate and heart rate from radio phase disturbances. It publishes the results as Home Assistant entities, Matter endpoints, or Apple HomeKit accessories. The whole stack claims to run edge-only: no camera, no phone app, no cloud subscription.

The interesting bit The project leans hard into the physics of multipath — your neighbors’ routers become free radar illuminators, and the system hops six WiFi channels to triangulate. A 4-bit quantized contrastive encoder fits in 8 KB and runs on a Raspberry Pi; the authors retracted an earlier “100% presence” claim and now report an 82.3% held-out temporal-triplet accuracy, which is either admirably honest or a red flag depending on your priors.

Key highlights

• 105 “Cog” edge modules for health, security, retail, and swarm use cases, loaded from a JSON registry
• Self-supervised pre-training on 60K frames, with optional MediaPipe-supervised fine-tuning in ~2 seconds on an RTX 5080
• ESP32-C6 support with 802.15.4 mesh time-sync and a ~5 µA low-power wake-on-motion mode
• Docker image available for simulation without hardware; PyPI wheels ship the Rust core via PyO3
• Ed25519 cryptographic witness chain for measurement attestation (unclear who verifies what)

Caveats

• The README is dense with ADR documents, benchmark claims, and integration badges; the actual wiring between ESP32 firmware, Seed hardware, and model inference is hard to trace
• “SOTA on MM-Fi” and other accuracy figures come from self-hosted benchmarks with no independent citation visible in the sources
• The $140 “total BOM” for a full Seed node is mentioned once without a parts breakdown

Verdict If you’re building privacy-first home automation or researching RF sensing, this is a feature-rich starting point with unusually broad smart-home integration. If you need reproducible research or a production safety system, the documentation sprawl and self-reported metrics warrant skepticism.