HTML explainer artifact

RuView

Source: https://github.com/ruvnet/RuView
Generated: 2026-05-22T06:48:30Z

RuView — WiFi CSI as camera-free spatial sensing

What this link is

GitHub repository: ruvnet/RuView, a Rust-first edge sensing platform that tries to turn WiFi Channel State Information (CSI) from ESP32-S3 / research NIC hardware into presence, vitals, motion, fall detection, multi-person counting, and experimental pose/spatial intelligence. The repo is MIT-licensed, public, and very active.

Core idea

Ordinary WiFi signals bounce through a room. Humans disturb those reflections through movement, breathing, posture changes, and body position. RuView reads CSI from low-cost WiFi hardware, transforms those streams into embeddings/features, then runs edge models and heuristics to infer room state without cameras or wearables.

Signal path:

ESP32-S3 CSI nodes / research NIC
  -> CSI stream + phase/amplitude features
  -> edge DSP / sensing server
  -> embeddings + heuristics + optional pretrained models
  -> room-level outputs: presence, breathing, heart-rate trend, motion, falls, occupancy, pose experiments
  -> dashboards / demos / optional Cognitum Seed cogs

Stack and architecture signals

• Primary language: Rust, with substantial JavaScript, Python, TypeScript, C, shell, and web assets.
• Hardware path: ESP32-S3 recommended for CSI; single-core ESP32/ESP32-C3 explicitly unsupported.
• Runtime modes: Docker/simulated data for evaluation, ESP32 live sensing for real data, optional Cognitum Seed appliance for persistent memory/vector store/witness-chain/edge cogs.
• Model assets: Hugging Face ruvnet/wifi-densepose-pretrained is referenced for CSI embeddings and presence detection; README also documents current format-loader gaps.
• Repo shape: large, fast-moving, with many ADRs, examples, demos, firmware, server pieces, and “cog” catalog references.

Maturity read

High activity and high public attention, but treat as beta/research-grade rather than plug-and-play product:

• README explicitly says beta software, APIs/firmware may change.
• Recent releases are automated and frequent; latest release notes include firmware watchdog fixes for ESP32-S3 PSRAM boards.
• Large surface area: firmware, edge server, web demos, pretrained weights, hardware provisioning, and optional appliance story.
• Claims are ambitious; README includes useful caveats, e.g. camera-free pose accuracy is still limited and camera-supervised training/evaluation phases are pending.

Install / try path

Best consumption order:

1. Read the README’s “What works today vs pending wiring” section before trusting demos.

2. Run Docker with simulated data first if evaluating UI/API shape.

3. Only then consider hardware: ESP32-S3 node(s), firmware flashing, WiFi provisioning, and local sensing server.

4. Use Hugging Face weights from Python/training tooling; README notes the live sensing-server model loader does not yet accept the JSONL RVF container, so the live path may need fallback mode until adapter work lands.

What is worth stealing for Dab / Mission Control

• “No camera, ambient sensing” is relevant to home/room awareness, but privacy and reliability need careful boundaries.
• The architecture pattern is useful: cheap edge sensors -> local feature extraction -> small specialized models -> privacy-preserving local dashboard.
• For Dab, the immediate useful angle is not installing this now; it is a research thread on whether Pi/ESP32 ambient sensing can support passive presence/occupancy signals for Mission Control.
• The README’s transparent gap table is a good documentation pattern for Dab project pages: separate demo claims, working paths, pending wiring, and hardware requirements.

Risk / skepticism notes

• WiFi CSI sensing can be environment-specific; calibration, node placement, multipath, and hardware quality matter.
• Vital signs and pose claims need independent validation before any safety/health use.
• The optional Cognitum Seed/edge-cog ecosystem is intertwined with the repo narrative; separate the open-source sensing stack from product/appliance claims when evaluating.
• Avoid treating GitHub stars or demo screenshots as proof of real-world reliability.

Suggested next action if Ananth cares

Create a Research item only if this becomes an active hardware exploration. First useful test would be a bounded spike: one ESP32-S3 CSI node + Pi/local server + simulated-vs-live comparison, with success measured as stable room presence/motion detection rather than vitals or pose.