radiom

A multi-source mobile web SDR client with a large built-in decoder library (88 protocols covering CW / RTTY / FT8/FT4 / WSPR / JT9/JT65 / FST4 / JS8 / NAVTEX / WEFAX / HFDL / ALE 2G / Olivia / MFSK / MT63 / Throb / ACARS / VDL-2 / ADS-B / AIS / DSC / DMR / NXDN / YSF / M17 / P25 / LRPT / APT / HRPT / LoRa / rtl_433 / SONDE / OP25 / AERO / STD-C / and ~50 others).

The browser app connects to one of three SDR-server flavors and routes the demodulated audio (or raw IQ, where appropriate) through a fleet of decoder bridges running on a small Node backend:

• KiwiSDR — source for HF (0–30 MHz) audio and IQ stream.
• OpenWebRX — second source supporting HF and VHF/UHF when the host has appropriate front-ends. Used for bands above 30 MHz when the operator has access to an OWRX-fronted receiver.
• rtl_tcp — third source for direct RTL-SDR access (typically via a remote RTL-SDR USB receiver exposed over TCP). Required for the IQ-baseband decoders that need ≥1 MS/s (ADS-B at 1090 MHz, UAT at 978 MHz, AERO / STD-C at L-band, LRPT / HRPT, OP25 trunking, etc.). rtl_tcp support is work-in-progress — known issues with audio/control coupling and tuning latency are being iterated on.
• Airspy SpyServer — fourth source, public Airspy R2/Mini/HF+ servers via the SpyServer binary protocol, with server-side csdr audio demod + waterfall synthesis on the radiom backend. SpyServer support is work-in-progress — audio quality varies with the upstream server's RF chain, the AGC chain still has artifact issues on some signals, and cursor/click-tune ergonomics are still being refined.

The active source is selected from the top bar; each decoder indicates in its tooltip whether it requires an IQ-capable source (rtl_tcp / OWRX) or works with the audio chain from any source.

Status: actively being tested. radiom is an advanced mobile-first receiver with a large surface area — many features still need real-world validation across receivers, bands, and propagation conditions, and more decoders / visualizers are on the way. Bug reports and on-air results are very welcome (see Getting in touch).

This README is a draft and does not list every shipped feature.

Screenshots

NOTE: The user interface has been significantly modified since the last screenshots.

See docs/SCREENSHOTS.md for a gallery of a small subset of the features.

Decoder library (88 protocols)

Alphabetical. Multiple decoders share a binary (e.g. jt9 covers four WSJT-X modes, dsd-fme covers ten digital-voice modes, multimon-ng covers twenty paging/selcall modes, direwolf covers four packet variants, inmarsat-sniffer covers AERO + STD-C).

Protocol	Vendored from	Variants / notes
9600 Packet	direwolf (MODEM 9600)	G3RUH scrambled NRZ, FOX cubesats + 70 cm — needs wideband (≥24 kHz) audio
ACARS	f00b4r0/acarsdec	131 MHz airline data link, MSK 2400 bps
ADS-B	flightaware/dump1090	1090 MHz Mode-S extended squitter
AERO	alphafox02/inmarsat-sniffer	Inmarsat AERO Classic (--mode=aero), L-band IQ
AFSK1200	EliasOenal/multimon-ng	Bell-202 AFSK 1200 bps (APRS, weather sondes…)
AFSK2400	EliasOenal/multimon-ng	Three tone-pair variants (AFSK2400 / _2 / _3) run concurrently
AIS	hessu/aisdecoder (gnuais)	161.975 / 162.025 MHz marine vessel tracking
ALE 2G	LinuxALE	MIL-STD-188-141A/B
AMTOR	fldigi navtex (B)	FEC mode of SITOR-B
APT	SatDump (noaa_apt)	NOAA analog APT @ 137 MHz
CCIR	EliasOenal/multimon-ng	ITU-R 5-tone selcall
CCITT	multimon-ng (→ CCIR)	ITU-T 5-tone — same tone table, routed via CCIR
CLIPFSK	EliasOenal/multimon-ng	Bellcore / ETSI Caller-ID V.23 FSK
Contestia (CTSA)	fldigi contestia	Same (tones × bandwidth) grid as Olivia
CSPAS	jbirby/COSPAS-SARSAT-Codec	406 MHz ELT / EPIRB / PLB (1G + 2G)
CW	from-scratch + fldigi	Single decoder, adjustable WPM bias
CWM	EliasOenal/multimon-ng	multimon-ng's native Morse (cross-check against CW)
D-STAR	lwvmobile/dsd-fme (-fd)	DV+DD digital voice, AMBE2+
DMR	dsd-fme (-ft)	ETSI Tier-2, single-slot
DMR-stereo	dsd-fme (-fs)	Decodes both TDMA slots concurrently (slot1=L, slot2=R)
DominoEX (DOMEX)	fldigi dominoex	DEX 4, 5, 7-8, 8, 11, 11-FEC, 16, 22
dPMR	dsd-fme (-fz)	dPMR446 (EU low-power business)
DSC	jbirby/DSC-Codec (Python)	ITU-R M.493 marine VHF Ch 70 + HF guard channels
DTMF	EliasOenal/multimon-ng	Touch-tone digits
DZ/PZVEI	EliasOenal/multimon-ng	German + Polish ZVEI dialects, bundled
EAS	EliasOenal/multimon-ng	Emergency Alert System SAME headers
EEA	EliasOenal/multimon-ng	European EAS-variant selcall
EIA	EliasOenal/multimon-ng	European industrial-alert selcall
EURO	multimon-ng (→ EEA)	Generic EU 5-tone, routed via EEA
FLEX	EliasOenal/multimon-ng	1600/3200/6400 bps 2/4-FSK paging
FLEX_NEXT	EliasOenal/multimon-ng	Newer FLEX revision, additive demod
FMSFSK	EliasOenal/multimon-ng	German FMS Funkmeldesystem 1200 bps
FreeDV	freedv_rx	700C / 700D / 700E / 1600 / 2020 digital voice
FSK9600	EliasOenal/multimon-ng	Generic 9600 bps NRZ FSK
FSQ	fldigi fsq	FSQ 1.5 / 3 / 4.5 / 6 baud
FST4	WSJT-X fst4d	TR 60 / 120 / 300 / 900 / 1800 s
FST4W	WSJT-X fst4d -W	TR 60 / 120 / 300 / 900 / 1800 s
FT4	from-scratch ft8/ft4	7.5-s slot decoder
FT8	from-scratch ft8/ft4	15-s slot decoder
HAPN4800	EliasOenal/multimon-ng	Hong Kong amateur packet 4800 bps FSK
HF Packet	direwolf (MODEM 300)	300-baud AX.25 / APRS on 30 m
HFDL	szpajder/dumphfdl	HF data link (multiple ground stations), IQ-mode
HRPT	SatDump (noaa_hrpt)	NOAA AVHRR HRPT @ L-band
IL2P Packet	direwolf (-d 2)	Nino Carrillo's Reed-Solomon FEC framing on VHF 1200
JS8	js8call	Normal, Slow, Fast, Turbo, Ultra
JT4	WSJT-X jt9 -4	sub-modes A–G
JT65	WSJT-X jt9 -65	—
JT9	WSJT-X jt9 -9	—
LoRa	tapparelj/gr-lora_sdr	EU 868 / US 915 / AS 433 MHz, SF7–12, BW 125/250/500 kHz
LRPT	SatDump (meteor_m2_lrpt)	Meteor M2 weather sat @ 137 MHz
LTR	GopherTrunk	LTR-Net trunked dispatch
M17	dsd-fme (-fU)	M17 open digital voice
MFSK	fldigi mfsk	MFSK4, 8, 11, 16, 22, 31, 32, 64, 128
MSK144	WSJT-X msk144d	Meteor-scatter
MT63	fldigi mt63	500S, 500L, 1000S, 1000L, 2000S, 2000L
NAVTEX	fldigi navtex	SITOR-B FEC broadcast (490 / 518 kHz / 4209.5 kHz)
NXDN-48	dsd-fme (-fn)	NXDN 4800 (narrow, Kenwood / Icom)
NXDN-96	dsd-fme (-fN)	NXDN 9600 (wider variant)
Olivia	fldigi olivia	18 (tones × bandwidth) presets: 4/125 … 64/2000
OP25	boatbod/op25 (rx.py)	P25 trunking + control-channel parsing, IQ-mode
P25-P1	dsd-fme (-fp)	P25 Phase 1 (CQPSK) — single-channel
P25-P2	dsd-fme (-f2)	P25 Phase 2 (HDQPSK) — single-channel
PACTOR	fldigi navtex (B)	Same engine, free-tuned dial
POCSAG	multimon-ng (POCSAG{512,1200,2400})	Pager protocol, all three baud rates
PSK (BPSK)	fldigi psk	PSK31, 63, 63F, 125, 250, 500, 1000, PSK125R/250R/500R/1000R
Q65	WSJT-X jt9 -q	Q65-A, B, C, D, E
QRSS	from-scratch (DSP)	Slow-CW visual decoder
RDS	windytan/redsea	FM broadcast Radio Data System
rtl_433	merbanan/rtl_433	~200 ISM-band protocols (weather, TPMS, meters, …)
RTTY	fldigi rtty	17 presets: 170 Hz amateur, 50/60/75/100 baud commercial, UK 200/50, Russian 200/100/250/75/450/75, DWD/TASS 425, 850/1000 Hz custom
SELCAL	EliasOenal/multimon-ng	Aviation HF 4-char SELCAL (-a CCIR)
SITOR-A	from-scratch	ARQ mode (full 7-char block)
SITOR-B	fldigi navtex (B)	FEC broadcast mode
SONDE	rs1729/RS (rs41mod et al.)	Radiosondes — RS41, DFM-09, M10, iMet-54, LMS6, MP3H1
STD-C	alphafox02/inmarsat-sniffer	Inmarsat-C SOLAS messaging (--mode=stdc)
THOR	fldigi thor	THOR 4, 5, 8, 11, 16, 22, 25×4, 50×1, 50×2, 100
Throb (THRB)	fldigi throb	Throb 1, 2, 4 + Throb-X variants
UAT	flightaware/dump978-fa	978 MHz US general-aviation ADS-B variant
UFSK1200	EliasOenal/multimon-ng	Universal FSK 1200 bps (telematics, telemetry)
VDL-2	szpajder/dumpvdl2	136 MHz aviation data link (12 channels concurrent)
VHF Packet	direwolf (MODEM 1200)	144.39 (US) / 144.80 (EU) MHz APRS, Bell-202
WEFAX	fldigi wefax	IOC 576 / 288, multiple LPM, B&W + colour
WSPR	WSJT-X wsprd	2-min slots, 11 standard sub-bands
WSPR-15	WSJT-X wsprd -m	15-min slots (LF / MF), aligned to :00/:15/:30/:45 UTC
WWV	fldigi wwv	WWV / WWVH minute-tick decoder
X10	EliasOenal/multimon-ng	X10 home-automation RF (310 MHz)
YSF	dsd-fme (-fy)	Yaesu System Fusion / C4FM
ZVEI	EliasOenal/multimon-ng	ZVEI1 + ZVEI2 + ZVEI3 selcall, bundled

Additional cross-cutting tooling:

• AGW Packet Engine server (TCP 8000) exposed by all four packet variants for APRSIS32 / UI-View / Xastir clients (RX-only).
• APRS telemetry (T# / PARM / UNIT / EQNS / BITS) decoded inline by all packet modes — direwolf's structured-payload parser is on.
• RSID auto-classifier vendored from fldigi for autonomous mode detection — runs alongside the AUTO panel and switches the active decoder.

THIRD_PARTY_LICENSES.md lists the upstream project each decoder is vendored from.

Viewer library (59 panels)

Real-time analysis views reachable from the VIEW picker. Tap a row to open the panel; long-press a row to read its detailed help overlay.

View	Input	What it shows
A/B Spectrum Compare	audio	Tap "A" to lock current Welch PSD as reference; display shows current (green), reference (orange), and difference in dB (white)
Allan Deviation Plot	IQ	σ_y(τ) of a PLL-tracked carrier on log-log axes (0.1–100 s) — receiver / source stability metric
Ambiguity Function	IQ	Full 2D |χ(τ, ν)| of the captured first-second reference waveform against itself; range-Doppler resolution map
Audio Constellation	audio	Quadrature-mixed demod audio plotted as a constellation; optional Costas lock (BPSK / QPSK / 8PSK)
Audio FFT	audio	16384-pt high-resolution audio spectrum 0–6 kHz with running average + hover cursor
Audio Scope	audio	Triggered oscilloscope on the demod audio, with auto / shot modes and adjustable trigger level / polarity
Audio Spectrogram	audio	Rolling waterfall of the demod audio with auto-stretch and contrast controls
Autocorrelation Function	audio	Real-time R(τ) over 0–50 ms lag; first peak gives the dominant period in ms / Hz
Bispectrum / Bicoherence	IQ	Third-order spectrum |B(f₁, f₂)| revealing quadratic phase coupling — nonlinear distortion, mechanical faults
Carrier Zoom	IQ	32k-sample long-FFT for sub-Hz resolution around the dial — read frequency drift to fractions of a Hz
Cepstrum / Pitch Contour	audio	Real cepstrum + scrolling pitch trace; voice F₀, MFSK tone spacing, echo delays
Channel-Sounder Echo Profile	IQ	Autocorrelation of |z(t)|² envelope vs lag τ (ms); multipath / skywave hop delays as peaks
Chromagram	audio	12-bin pitch-class profile (C, C#, …, B) folded across octaves; tonal-centre / chord tracking
CI Periodogram	audio	Welch PSD with 95% χ² confidence-interval band; identifies which spectral peaks are statistically significant
Clock Drift	IQ	Parts-per-million offset of the receiver clock vs a stable reference carrier, with strip chart
Constellation Density Heatmap	IQ	2D log-scaled histogram of IQ samples; high-SNR clusters bright, noise diffuse
Crest Factor / PAPR	audio	Peak/RMS ratio in dB strip-chart — ~3 dB sinusoid, ~10 dB noise, ≥15 dB impulsive interference
Cyclic Spectrum SCF	IQ	|S(f, α)| in spectral frequency × cycle frequency plane; reveals symbol-rate fingerprints
Cyclostationary CAF Heatmap	IQ	|R(α, τ)| over τ ∈ {4…512} samples; symbol rates of digital modes (PSK31, RTTY, FT8, MT63…) appear as bright stripes at α = baud
Delay-Doppler Scattering	IQ	2D channel scattering map (delay × Doppler) for ionospheric / multipath diagnosis
Doppler Tracker	IQ	PLL-locked Δf of the strongest in-band carrier vs time, 30-min strip
Envelope PDF	audio	Probability density of the analytic envelope |x + jH{x}|; Rayleigh / Rician / bimodal / speech shape with heuristic classifier
EVM / MER vs Time	IQ	Costas-locked QPSK constellation; Error Vector Magnitude % + Modulation Error Ratio in dB strip with link-quality reference lines
Eye Diagram	IQ	Symbol-period folded IQ trace — eye open = clean digital reception, eye closed = noise / clock mismatch
GPSDO Lock / Holdover	IQ	Cumulative phase of PLL'd carrier over panel session; slope = drift in Hz/s, jumps = unlock events
Group Delay vs Frequency	audio	τ_g(f) = −dφ/dω of the receive chain; measures filter ringing and all-pass equalisation
Higher-Order Cumulants	IQ	Live |C₂₀|, |C₂₁|, |C₄₀|, |C₄₂|, μ₄₂ traces with auto-classifier (noise / BPSK / QPSK / 16-QAM / FM)
Hilbert-Huang Transform	audio	EMD-extracted IMF1/2/3 instantaneous-frequency strip charts; adaptive multi-component time-frequency analysis
Instantaneous Frequency Trace	IQ	d/dt(arg z) on raw IQ as a 4-s scrolling trace; FM voice, chirped radar, drift slope all visible
I/Q Imbalance Monitor	IQ	Amplitude imbalance (dB) / phase imbalance (°) / DC_I / DC_Q time series — SDR front-end diagnostic
IQ Constellation	IQ	Raw I-vs-Q scatter of complex baseband
Kurtosis vs Time	IQ	4th-order moment of IQ amplitude — Gaussian noise sits at 3, sub-Gaussian (carrier) below, super-Gaussian (bursts) above
LPC Spectral Envelope	audio	Levinson-Durbin order-12 all-pole envelope with formant markers (F1 / F2 / F3 / F4)
Mains-Hum Tracker	audio	Narrow-band FFT + scrolling waterfall over 0–250 Hz with reference lines at 50 / 60 / 100 / 120 / 150 / 180 / 200 / 240 Hz
Mel Spectrogram	audio	64 Mel-spaced filter banks 100 Hz–6 kHz; auditory-perceptual time-frequency map
Mirror-Image Rejection	IQ	Positive-band vs negative-band power ratio (dB) over time; image-suppression metric
Modified Allan / TDEV	IQ	MDEV + HDEV + TDEV + ADEV on log-log axes; the full Allan-variance family
Modulation Spectrum	audio	STFT-of-STFT — modulation-rate × acoustic-frequency heatmap; reveals AM/FM rate, speech syllable rate (3–8 Hz)
Over-The-Horizon Radar	IQ	Slow waterfall + FMCW / pulsed-sweep classifier for OTH radars (Container / ROTHR / JORN / …)
Phase Noise Spectrum L(f)	IQ	Single-sideband phase-noise PSD in dBc/Hz vs offset frequency from PLL'd carrier; standard oscillator metric
Phase Portrait	audio	Trajectory in (x(t), x(t+τ)) embedding space with adjustable τ slider; nonlinear-dynamics classic
Q-Q Plot	audio	Quantile-quantile envelope vs Rayleigh / Gaussian / Rician theoretical distribution; goodness-of-fit
Range-Doppler Map	IQ	Radar-style 2D detection plot — range (delay) × Doppler (velocity proxy); first-second reference + cross-correlation
Reassigned Spectrogram	audio	STFT with bin-by-bin time-frequency reassignment via phase-derivative window operators; sharper than ordinary spectrogram
RFI Visualizer	IQ	Catalogues narrow tones detected in the IQ window over time — freq, magnitude, last-seen
Sferic Lightning Visualizer	IQ	Counts wideband impulse spikes per second; rolling 60-s bar chart with peak / floor readout
Signal Meter	RSSI	Analog-style S-meter dial (dBm and S-units)
Signal over Time Plot	RSSI	RSSI strip chart, last 60 s + full-session capture/export
Spectral Coherence γ²(f)	IQ	Magnitude-squared coherence between two consecutive IQ time windows; γ²≈1 = stationary, γ²≈0 = transient
Spectral Entropy	audio	Shannon entropy of normalised PSD strip chart; 0 = pure tone, 1 = white noise; classifier with class labels
Spectral Flatness	audio	Wiener entropy (geometric/arithmetic mean ratio of PSD); standard tonal-vs-noise discriminator
Spectral Mask Overlay	audio	Welch PSD with cycleable regulatory transmit mask (SSB 2.7 / AM 6 / NBFM 12.5 kHz / Flat)
Stockwell S-Transform	IQ	Multi-resolution time-frequency map with σ ∝ 1/f Gaussian window — wavelet resolution + STFT phase reference
Symbol Clock Recovery Scope	IQ	PLL on the symbol-clock spectral line; plots tracked baud + phase error and reports lock state
Vector / Lissajous Plot	audio	x(t) vs x(t−τ) on demod audio with adjustable τ (1–3000 samples); reveals modulation symmetry / harmonics
Voice Formant Tracker (exp)	audio	Tracks the first three vocal-tract resonances (F1 / F2 / F3) on speech audio
Wavelet Scalogram	audio	Morlet continuous wavelet transform at 64 log-spaced scales (30 Hz – 4 kHz); constant-Q resolution
Welch PSD	audio	Averaged-periodogram PSD with persistent min-hold floor trace; cleaner noise-floor measurement than Audio FFT
Wigner-Ville (SP-WVD)	IQ	Smoothed-Pseudo Wigner-Ville Distribution; quadratic time-frequency with cross-term suppression

Integrated signal generator (GEN button)

Some shipped decoder can be exercised without a live signal via the GEN button in the function row. It opens a categorised picker (openModesPicker) of curated test samples — synthesized reference transmissions plus a handful of real off-air recordings — covering CW, RTTY (every shift / baud combination listed above), all BPSK / QPSK / 8PSK rates, every MFSK / Olivia / Contestia / MT63 / DominoEX / THOR / Throb variant, FSQ, FT4 / FT8 / FST4 / WSPR / JS8 / JT9 / JT65, NAVTEX, WEFAX, Hellschreiber, ALE, packet, IFKP, and WWV.

Quick start

npm install
npm run dev      # HTTPS Vite dev server (uses mkcert) — for phone testing
npm run dev:http # plain HTTP on :5174 — for laptop / headless browsers
node server.mjs  # decoder backend on :8080 (Vite proxies /ws/decode/* here)

Decoder binaries are built per-decoder via npm run build:<name> (see package.json). The full set is also assembled by the multi-stage Dockerfile used for the fly.io deployment.

Project layout

src/             — TypeScript browser app (UI, audio graph, decoder clients)
public/          — static assets + AudioWorklet processors
decoder/         — Node-side decoder bridges (spawn binaries, forward over WS)
decoders/        — per-decoder build scripts and source glue
  decoders/<x>/build.sh   # clones upstream, builds binary into bin/
  decoders/<x>/main.cpp   # glue between fldigi/upstream API and our stdio protocol
server.mjs       — HTTP + WebSocket server, hosts /ws/decode/* endpoints
vite.config.ts   — dev-server config + /ws/decode proxy to server.mjs
Dockerfile       — multi-stage production build (Debian + Node 22)

Deployment

The canonical deploy target is fly.io. fly.toml is committed and assumes a shared-CPU VM with 512 MB RAM in region yyz (Toronto) — fork and edit those values for your own app/region.

fly launch --no-deploy   # only the first time, to claim an app name
fly deploy               # build the multi-stage Docker image and ship it

For self-hosters who don't use fly, the Dockerfile is standalone:

docker build -t radiom .
docker run --rm -p 8080:8080 radiom

The image exposes port 8080 and runs node server.mjs, which serves both the static front-end and the /ws/decode/* decoder bridges. All decoder binaries are built inside the image, so the runtime container needs no compilers — just a few shared libs (libasound2, libcodec2, libpng16-16, libsndfile1).

Backend auth

All WebSocket endpoints (/ws/* — both /ws/decode/* decoder bridges and the /ws/<host>:<port>/... Kiwi audio/waterfall proxy) accept an optional shared bearer token plus an Origin allow-list, both controlled by env vars on the server:

Env var	Default	Effect
RADIOM_TOKEN	(empty)	Required token in ?token=… query string on all /ws/* upgrades. Empty = open.
RADIOM_ALLOWED_ORIGINS	(empty)	Comma-separated. Empty = any origin. Entries beginning with . match suffixes (e.g. .fly.dev).
RADIOM_MAX_WS_PER_IP	4	Max concurrent decoder WS per client IP.
RADIOM_MAX_WS_GLOBAL	32	Max concurrent decoder WS across the whole server.

On fly.io, set the token via:

fly secrets set RADIOM_TOKEN=$(openssl rand -hex 24)
fly secrets set RADIOM_ALLOWED_ORIGINS=https://your-app.fly.dev

The browser-side token lives in localStorage under radiom.token. Users either type it into Settings → Backend Access → Server token, or open a share-link like https://your-app.fly.dev/#token=… and the token gets stored

• stripped from the URL on first load.

Getting in touch

• Issues — bug reports, feature requests, decoder problems, build failures.
• Discussions — open questions, "how do I…", on-air results, ideas you want to talk through before opening an issue.
• Security — for anything sensitive, please use GitHub's private security advisories rather than a public issue.

Contributing

radiom is a personal project, open-sourced so others can read the code, and fork it for their own use. Pull requests are not accepted — the codebase stays solo-authored. Bug reports, on-air results, and ideas are very welcome through Issues and Discussions; anything that lands in the upstream codebase will be implemented from there.

Authorship

Every line of application code in this repository was written by Claude Code (Anthropic) under the direction of Andre Paquette — design decisions, decoder selection, and on-air validation were human-driven; the actual TypeScript, C/C++ glue, build scripts, and server code are the assistant's work.

Third-party decoder sources vendored under decoders/<name>/ retain their original authorship and licensing (see THIRD_PARTY_LICENSES.md).

License

radiom is released under the GNU General Public License v3.0 or later — see LICENSE for the full text and THIRD_PARTY_LICENSES.md for the upstream projects this repository builds against (fldigi, wsjt-x family, multimon-ng, direwolf, dumphfdl, slowrxd, JS8Call, LinuxALE, …).

The project ships cherry-picked fldigi DSP/protocol source in-tree and links against several GPL-2/GPL-3 decoder binaries at runtime, so the combined work must be distributed under GPL-3.0-or-later.

Copyright © 2026 Andre Paquette and radiom contributors.

OpenAI Whisper integration

The live-transcription feature uses OpenAI's Whisper API. Your API key is entered through Settings (cog menu) and stored only in browser localStorage on your device — it is never committed to source and never transmitted anywhere except to api.openai.com over HTTPS when a transcription request is made.