A personal, edge-computed EEG + HRV biofeedback platform — built by one person over several years,
self-hosted on a Raspberry Pi 4, and open-sourced for anyone who wants their physiology data
to stay theirs.
neurovis.io ·
About / Contact ·
Research & References
Neurovis combines real-time EEG (Muse headband) and HRV (Polar H10 chest strap / Apple Watch) sensor data with AI-assisted analysis, built around one core idea: processing should happen locally, and your biometric data shouldn't have to live on someone else's server to be useful. Where a server is involved at all, it's the author's own Raspberry Pi, not a cloud platform.
This repo isn't a single app — it's three related tools that grew organically over the project's life. Each is documented in its own section below.
 |
 |
Live Dashboard — real-time valence/arousal, HR/HRV, and Muse/Polar sensor status (Fusion/Neurovis.html) |
Continuous Forensics — longitudinal trend reports across all logged sessions (Fusion/MindRelay.html) |
 |
 |
Session Deep-Dive — high-density forensic replay of a single session (Fusion/emotionreplay.html) |
HRV Dashboard — Apple Watch sleep/HRV/workout trends (HRV/NeurovisAW.html) |
| Directory | What it is | Runs where |
|---|---|---|
Fusion/ |
The real-time dashboard: connects directly to a Muse EEG headband and Polar H10 chest strap over Web Bluetooth, and runs all the signal processing in-browser via Pyodide (Python compiled to WebAssembly). No backend process is involved — this is the architecture that actually delivers on the "your data stays local" promise. Session history is kept in the browser's IndexedDB. | Static files, any browser |
Agents/ |
A local FastAPI + LlamaIndex multi-agent chatbot for conversational analysis of exported session/health data — a router agent delegates to a meditation/EEG specialist and an Apple Watch/HRV specialist. Runs against a local Ollama model (qwen2.5), not a cloud LLM. Not currently exposed to the public web — it's a personal analysis tool. |
Locally, via python3 |
HRV/ |
Companion pieces for Apple Watch data: NeurovisAW.html is the standalone HRV dashboard (screenshot above), and uploader.py is a small Flask endpoint that optionally accepts an anonymized, opt-in daily summary for research purposes — nothing leaves the browser unless you explicitly click "Upload to Neurovis." |
uploader.py on the Pi; HTML anywhere |
There's no build step, package manifest, or bundler — everything here is plain Python/HTML/JS, run directly.
Pure static files. Serve the directory and open it in a browser (Web Bluetooth requires Chrome/Edge and HTTPS or localhost):
cd Fusion && python3 -m http.server 8080
# open http://localhost:8080/Neurovis.htmlPyodide fetches hardware.py and Neurovis.py from this same server and runs them in-browser at ~20Hz — there's nothing else to configure.
Requires a local Ollama install with the qwen2.5 model, plus:
pip3 install fastapi uvicorn pandas numpy scipy matplotlib pydantic \
llama-index-core llama-index-llms-ollama llama-index-embeddings-huggingface
ollama pull qwen2.5
cd Agents && python3 neurovisAgent.py
# open http://localhost:8000Upload an Apple Health export and/or a Neurovis session CSV/JSON through the UI, then chat with it — the router sends your question to whichever specialist agent has the relevant data loaded.
cd HRV && python3 -m http.server 8081
# open http://localhost:8081/NeurovisAW.html, then Import Data / Upload Zip
# optional: the anonymized-upload receiver
pip3 install flask flask-cors
python3 uploader.py # binds :8002This is a solo, ongoing personal project, not a medical device or clinical tool — metrics like the Anxiety/AXX scores, Functional Beta Asymmetry, and similar are experimental research constructs, not diagnostic measures. Some pieces (EEG/ECG live recording in particular) are explicitly alpha quality per the site itself. Use it, learn from it, fork it — just don't treat it as medical advice.