I built a tool that syncs 25+ BLE smart scales to InfluxDB, Home Assistant, Garmin, Strava, and more, no phone app needed

#opensource #iot #typescript #selfhosting

The problem

I own a cheap BLE smart scale. It measures weight and body composition (fat %, muscle, water, bone mass…) over Bluetooth. Solid hardware. The catch: the only way to get those numbers into Garmin Connect was this ritual:

Step on the scale.
Open the vendor's phone app.
Wait for it to connect and sync.
Read the numbers off the app.
Manually type them into Garmin Connect. Every. Single. Time.

If I forgot my phone, or the app's cloud was down, or I just couldn't be bothered — no data point. For something that's only useful as a trend, missing days quietly kills the whole point.

I didn't want to depend on a phone app that talks to a vendor cloud I don't control. So I built BLE Scale Sync.

What it does

A small always-on device (a Raspberry Pi Zero 2 W in my case) sits next to the scale and listens over Bluetooth. You step on, wait a few seconds, step off — and the reading lands in Garmin Connect. No phone. No app. No manual entry. No vendor cloud. Your data stays on your hardware.

Garmin is just one target. The same measurement fans out to whatever you configure:

Garmin Connect
Strava
MQTT → Home Assistant (auto-discovery, sensors appear automatically)
InfluxDB → Grafana dashboards
Webhook → anything
Ntfy → push notification on your phone
File → CSV / JSONL for your own scripts

And it speaks 25+ scale brands — Xiaomi (Mi Scale 2 passive broadcast), Renpho / FITINDEX / QN-Scale, Eufy, Yunmai, Beurer, Sanitas, Medisana, and more. Many protocols were ported from the excellent openScale project; a handful I reverse-engineered for this one.

Setup is one command

There's an interactive wizard that does BLE scale auto-discovery, user profiles, exporter selection, and connectivity tests for you:

docker run --rm -it --network host --cap-add NET_ADMIN --cap-add NET_RAW \
  --group-add 112 -v /var/run/dbus:/var/run/dbus:ro \
  -v ./config.yaml:/app/config.yaml \
  ghcr.io/kristianp26/ble-scale-sync:latest setup

It generates a config.yaml you never have to touch by hand, but it's plain and readable if you want to:

ble:
  scale_mac: 'FF:03:00:13:A1:04'   # so you don't grab a neighbour's scale

exporters:
  garmin:
    enabled: true
  mqtt:
    enabled: true
    broker_url: 'mqtt://homeassistant.local:1883'

Then run it continuously and forget it exists:

docker run -d --restart unless-stopped --network host \
  --cap-add NET_ADMIN --cap-add NET_RAW --group-add 112 \
  --device /dev/rfkill -v /var/run/dbus:/var/run/dbus:ro \
  -v ./config.yaml:/app/config.yaml:ro \
  -e CONTINUOUS_MODE=true \
  ghcr.io/kristianp26/ble-scale-sync:latest

If you run Home Assistant OS/Supervised, it's a one-click add-on instead — UI config, Mosquitto auto-detected, Garmin tokens bootstrapped on first start.

A few things I'm proud of

No Pi next to the scale? Use a $5 ESP32. Bathrooms are often out of Bluetooth range of wherever your server lives. So there's an ESP32 BLE proxy: it sits by the scale and relays BLE over WiFi/MQTT to the container running anywhere on your network. It even ships an embedded MQTT broker so there's zero broker setup. If you already run an ESPHome Bluetooth proxy mesh for Home Assistant, it can reuse that instead (multi-proxy with automatic RSSI-based pick).

Multi-user. Two people, one scale. It identifies who stepped on by weight range and routes to per-user exporters (my Garmin, your Garmin).

Historical backfill. Some scales cache offline measurements. On reconnect it replays them with their original timestamps to targets that support back-dating (Garmin, InfluxDB, File), so a week away from the Pi doesn't leave a hole in your chart.

Linux BLE is… a journey. BlueZ has a notorious "stuck discovery" state where scanning silently dies after hours/days. There's a consecutive-failure watchdog that auto-recovers it, plus optional systemd Type=notify integration for whole-loop freezes. Headless reliability was honestly the hardest part — way harder than parsing the scale protocols.

Body composition is computed locally. Scales send weight + raw bioelectrical impedance; the 10 body-composition metrics are derived on-device via BIA formulas. Nothing leaves your network.

Stack

TypeScript / Node 22, zod for config validation, node-ble (Linux) / noble (macOS/Windows) for BLE, mqtt + an embedded aedes broker, Vitest. Adapters and exporters are both pluggable registries — adding a new scale or export target is a single self-contained module, which is documented in CONTRIBUTING.md if you want to add yours.

Cross-platform: Linux (Docker + native), macOS, Windows. GPL-3.0.