An offline Android app that demonstrates covert data transmission — hiding data in sound and images — alongside the detection techniques that can catch it.
nightjar pairs four data-hiding techniques with their detection/steganalysis counterparts, all running locally on-device:
- Acoustic data-over-sound modem — sends data phone-to-phone, speaker to microphone, room-local, using FSK/PSK/MFSK encoding.
- Image steganography — least-significant-bit (LSB) encoding of data inside images, plus a steganalysis screen for detecting it.
- Audio steganography — four techniques (phase inversion, spectrogram-LSB, MFSK, and true phase-coding) for hiding data inside an audio clip.
- Passive acoustic detector — a real-time listener that can flag the modem's own signal, demonstrating that "covert" acoustic channels aren't necessarily undetectable.
The default UI is "Night Jar," a disguise-themed front end where every decoded payload appears as a firefly. The underlying technical module screens (modem, image stego, audio stego, detector) are reachable via a long-press reveal gesture from the jar screen.
Real screenshots from a Pixel 6a. The app opens as Night Jar — a calm, disguise-themed front end where every hidden payload shows up as a "firefly." A long-press on the jar reveals the technical Workshop underneath.
 Welcome first-run trail |
 Night Jar the home shelf |
 Inside a jar your fireflies |
 In sound acoustic modem |
 In an image LSB bit-plane |
 In audio spectrogram diff |
 The Meadow passive detector |
 The Workshop module menu |
 Image tool embed · extract |
Grab the latest signed APK from the Releases page
and install it directly — no Play Store needed. On your phone: download the
.apk, open it, and allow "install unknown apps" for your browser or file
manager if prompted.
Prefer to track updates automatically? Obtainium
can watch this GitHub repo and notify you of new releases — add
herakles-dev/nightjar as a GitHub source.
- Android 12+ (minSdk 31), target/compileSdk 35
- A single permission:
RECORD_AUDIO(used by the acoustic modem and detector to capture microphone input) - Android SDK + JDK 17 to build
./gradlew assembleDebug # build debug APK (output: app-debug.apk)
./gradlew test # run the JVM unit test suite636 JVM unit tests cover the signal-processing core: the acoustic modem and detector, image and audio steganography round-trips, Reed–Solomon error correction, the LSB bit-plane and steganalysis capacity sweeps (including a false-flag check against real photographs), and the incoming-payload pipeline. CI runs Android lint, the test suite with coverage, and a debug build on every push and pull request, and CodeQL and OpenSSF Scorecard run alongside it.
| Check | Result |
|---|---|
| Android lint | 0 errors (CI-enforced) |
| Line coverage, signal-processing core | 85.6% (2,654 lines) |
| Line coverage, whole app | 33.2% — the Compose UI screens are not unit-tested |
| Mutation score, modem / Reed–Solomon / FFT / spectrogram / detector | 89.8% (574 of 639 mutants killed) |
Coverage says a line ran; mutation testing (PIT) says a test would notice if the line were wrong. Running it exposed a weaker spot in the acoustic detector (69.5% killed) that new exact-output tests lifted to 85.3%. To reproduce:
./gradlew lintDebug jacocoTestReport # lint + coverage: app/build/reports/
./gradlew pitest # mutation tests, ~15 min: app/build/reports/pitest/Mutation testing also runs weekly in CI. Scope is the JVM-only core; Robolectric and Compose code is too slow to mutate meaningfully.
Kotlin + Jetpack Compose + Room, built with AGP 8.7.3 / Kotlin 2.0.21. Each
technique lives behind a shared CovertModule interface, with its
transmit/embed logic in a *Carrier class and its detection logic in a
matching *Detector/*Steganalysis class:
modules/fireflyjar Default disguise-themed UI (jar shelf, firefly log, player)
modules/acoustic Acoustic FSK/PSK modem screen
modules/audiostego Audio steganography screen (embed/extract)
modules/imagestego Image LSB steganography screen
modules/detector Passive acoustic detector screen
picker Technical module-picker home screen
trail Onboarding "riddle trail" walkthrough
share Outbound sharing (FileProvider-based)
incoming Unified incoming-payload pipeline: sniff, route, decode
Root-level classes (AcousticCarrier, AudioStegoCarrier, ImageStegoCarrier,
SturdyImageCarrier, AcousticDetector, AudioStegDetector,
ImageSteganalysis, Fft, Spectrogram, WavFile, MicCapture, …)
implement the signal-processing and steganalysis logic that the module
screens share.
nightjar is a research and educational demonstration of steganography and covert-channel techniques, and of the detection methods used against them. It is meant for learning, teaching, and experimentation in controlled settings — not for evading lawful monitoring or transmitting data without the consent of everyone involved. Please use it responsibly and lawfully.
nightjar's techniques are directly inspired by, and in some cases adapt, the published work of others — most notably Benn Jordan's acoustic data-hiding projects (Wavest, BUM16, AlphaSteg) and Georgi Gerganov's ggwave, along with foundational steganography and audio-watermarking research. See CREDITS.md for full acknowledgments.
MIT — see LICENSE. Third-party components (ggwave, the Silkscreen font) are used under their own licenses; see NOTICE for details.
Issues and pull requests are welcome.