// Real GibberLink Protocol · ggwave Engine · 100% Offline · No Files Needed
The actual data-over-sound protocol that went viral when two AI agents detected each other and switched mid-conversation. Encode English into FSK audio. Decode it back. Works over any speaker and mic.
// English → GibberLink Sound
// GibberLink Sound → English
Drop a .wav here, or
Accepts .wav files containing GibberLink audio
Point your mic at a GibberLink audio source. Decoding is automatic — the result appears the moment a valid transmission is detected.
// Protocol Details
Multi-frequency Frequency-Shift Keying. 3 bytes transmitted simultaneously via 6 tones across a 4.5 kHz band split into 96 frequencies. dF = 46.875 Hz per step.
Built-in error correction recovers data from noisy or partially clipped audio. Tolerates real-world interference, background noise, and distance.
No network required. Works between completely isolated devices using only speakers and microphones. Ideal for offline pairing and IoT provisioning.
Audible Fast: ~16 bytes/sec. Ultrasound modes run at 15–19.5 kHz — inaudible to most humans, but still captured by nearby microphones.
Built by Boris Starkov and Anton Pidkuiko. Won first place at the ElevenLabs Worldwide Hackathon in 2025. Two AI agents switched mid-conversation — 15M+ views on X.
This page uses the real ggwave engine by Georgi Gerganov — compiled to WebAssembly and fully embedded. No external files, no server, works offline. MIT licensed.