A local-first desk companion that watches your posture, paces your focus, and nudges you to stay healthy — with everything running on-device.
Note
Stillpoint runs 100% on-device — your webcam feed never leaves your computer, and there are no accounts, servers, or telemetry.
Stillpoint is a desktop app that helps you sit better and work healthier without ever sending your camera anywhere. It tracks your head and shoulders through your webcam with on-device pose detection, scores your posture against a personal calibrated baseline, and gently nudges you when you slouch — then surrounds that with a Pomodoro focus timer, a task and sticky-note workspace, and opt-in wellness reminders. The whole thing is private by design: the pose model is bundled and runs locally, so the only thing stored is a handful of numbers (your baseline, settings, and analytics) on your own machine.
Download the latest Windows installer from the latest release →:
- Windows 10/11 —
Stillpoint-x.y.z-setup.exe
A webcam is required; no GPU needed.
Important
The build is unsigned (no paid code-signing certificate), so on first launch Windows SmartScreen may warn you — click More info → Run anyway. Then allow camera access and calibrate from the Dashboard.
Note
macOS & Linux: Stillpoint is built on cross-platform Electron and runs on macOS and Linux when built from source (below). Prebuilt macOS/Linux downloads aren't provided — a free, unsigned macOS app can't launch cleanly on Apple Silicon without a paid Apple Developer ID, so a Windows installer is the only official download.
🛠️ Build from source (Windows · macOS · Linux)
# Requires Node.js 20 LTS + a webcam
npm install
npm run dev # run in development
npm run build && npm start # preview the production build
npm run build:win # package a Windows installer → dist/
# (npm run build:mac / build:linux also exist for source builds) |
|---|
 |
 |
|---|---|
 |
 |
- On-device pose tracking — a bold skeleton overlay on your live camera feed, powered by MediaPipe; no GPU required.
- 10-second guided calibration — captures your neutral posture as a personal baseline; recalibrate any time.
- Three issues, individually toggleable — forward head, shoulder slouch, and shoulder tilt; turn off what you don't want scored.
- Smoothed 0–100 score — Low / Medium / High sensitivity presets.
- Sustained-bad-posture alerts — configurable 15–120s threshold, desktop notification + sound, optional repeat, and a positive confirmation when you correct.
- Presence-aware — auto-pauses when you step out of frame, turn away, or the camera is covered or too dark, so absence never counts against you.
- Pause/resume + live streaks — track your current and best good-posture streak for the session.
- Pomodoro timer — configurable focus / short-break / long-break / cycles, auto-start, a progress ring, and phase notifications.
- Tasks & sticky notes — a quick task list and colored, inline-editable notes.
- Opt-in reminders — movement break, hydration, eye care (20-20-20), stretch, and posture reset, each with its own interval and live countdown. A long Pomodoro break even suggests a stretch and resets those reminders.
- Local analytics — daily good/bad percentages, issue breakdown, alerts, corrections, focus sessions, and a switchable trend chart (good posture % / focus sessions / best streak). Export to JSON any time.
- Five themes — Midnight, Nord, Graphite, Aurora, and a light Daylight, applied instantly.
- Keyboard shortcuts —
Spacestart/stop ·Ppause ·Ccalibrate ·1–6switch tabs — plus toast feedback and accessibility focus rings.
All posture processing runs locally on-device. No webcam frames or video are uploaded or written to disk. Stillpoint stores only local configuration, your calibration baseline, and local analytics summaries — all on your own machine.
A full reference manual lives in docs/ — architecture, features, services, data model, design system, and the release process.
Anahat Mudgal
- Website: anahatmudgal.com
- GitHub: @AnahatM
- Project Page: anahatmudgal.com/projects/stillpoint
This project is open source and available under the MIT License.