Breaking

Tether Open-Sources BrainWhisperer, an On-Device Brain-to-Text Engine, Inside Its QVAC AI Platform

July 16, 2026 at 14:01 EDT

Open Source
Healthcare
Foundation Models

Tether has fully open-sourced BrainWhisperer, a brain-to-text engine that decodes neural signals into words entirely on-device, integrating it into the company's local-first QVAC AI platform. Chief executive Paolo Ardoino announced on July 16, 2026 that the model is now 100% open source and available to developers.

July 16, 2026 · Tether

BrainWhisperer: Brain-to-Text Decoding, Now Fully Open Source

Tether has released its neural-signal-to-text engine as 100% open source, running entirely on-device in under 2GB — so sensitive brain data never leaves the user's hardware.

98.3%

accuracy on the benchmark task

1.78%

word error rate

<2GB

local footprint, fully on-device

4th

of 466 teams in Brain-to-Text '25

Accuracy, drawn to scale

Correctly decoded words vs. errors — out of every 100 words

98.3%

Correct

1.78%

Errors

How on-device decoding works

Neural signals

recorded from device

→

Whisper-based backend

bci-whispercpp, on-device

→

Word-by-word text

data never leaves device

Why it's welcomed

Privacy-first: inference stays on the user's hardware
Open source and free for developers to build on
Could restore text communication for those who've lost speech

The caveats

Still an early proof of concept
Real-world developer evaluations remain limited
Clinical or consumer deployment needs more work

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.