SuperFind

Find a device by its radio. Locate Bluetooth things in real space, with honest uncertainty instead of false precision.

Rust core Android + Linux CLI No internet Open source v0.1.0

Download View source

No internet permission. SuperFind does all of its sensing and maths on-device, so nothing it hears over the air can leave your machine. Bluetooth scanning is declared non-location, and rotating MAC addresses are never persisted.

What it does

SuperFind helps you physically locate a Bluetooth device, a tag, an earbud, a speaker, using nothing but its radio signal. Signal strength is a coarse proxy for distance, so instead of pretending to know exactly where something is, SuperFind runs a particle-filter fusion engine that tracks the target across noisy, often ring-shaped evidence and shows you a confidence region rather than a fake pinpoint. The same Rust core powers a Linux command-line tool and an Android app.

Key features

Device discovery

Lists nearby BLE devices by their broadcast names instead of rotating hex MAC addresses.

Hunt mode

Target one device and get a live signal readout and distance estimate as you move toward it.

Radar view

A swept aperture that remembers the sectors you have already scanned to guide your search.

Honest map

Draws a 95% confidence ellipse rather than a single point, so the uncertainty is visible, not hidden.

Calibration

Per-device path-loss fitting at 1, 2, 4 and 8 metres, with physical-plausibility checks on the fit.

One core, many platforms

An identical fusion core runs the CLI and the Android app, with more platforms planned.

Screenshots

Android app

Command line (Linux)

How it works

Radio samples flow through a platform-agnostic core that keeps full uncertainty at every step, so the UI can render honest confidence rather than a single guess.

observe (RSSI)→ path-loss→ particle filter→ motion model→ bearing→ tracker snapshot

A particle filter is used instead of a Kalman filter because the likelihood from a single distance reading is ring-shaped and often bimodal, which a Gaussian cannot represent. Motion comes from pedestrian dead reckoning (typed on the CLI, sensed on Android), and the bearing estimator returns "unknown" rather than a low-confidence guess. Confidence is the product of coverage, concentration and significance, multiplied rather than averaged, so any one of them being near zero sinks the answer.

Built with

Rust (zero-dependency core)KotlinJetpack Compose JNI bridgezbus / BlueZ (D-Bus)Cargo workspace 115 tests

Download

Current release: v0.1.0. Verify against checksums.txt from the release if you like.

Android app

Compose UI with radar and confidence-ellipse map. Verified on a Moto G40 Fusion, Android 6 to 16.

Debug-signed APK Download APK

Linux CLI

Terminal survey and hunt modes over BlueZ. No root needed, just a running bluetoothd.

x86_64 binary Download CLI

Early release. The APK is debug-signed, so on install Android may warn about an unknown source; that is expected. On Linux, make the binary executable (chmod +x) and run it with Bluetooth powered on.

Discussion

Tried it, or curious how the fusion works? Leave a comment.