Zuna open-source thought-to-text model now official

Zuna is an open-source 380‑million‑parameter foundation model built for EEG (electroencephalogram) signals, positioning itself as a first step toward reliable thought‑to‑text systems. EEG uses scalp electrodes to capture the brain’s electrical activity, but the resulting signals are typically noisy, inconsistent between subjects, and sometimes missing channels entirely.

Rather than directly decoding thoughts into language today, Zuna focuses on denoising and reconstructing EEG data. It cleans up raw signals, fills in missing channels, and predicts higher‑resolution activity from lower‑quality inputs, standardizing the data for downstream models.

The long‑term vision is for such foundation layers to sit beneath future decoders that can map cleaned EEG patterns to natural language. Zuna’s weights (under 2 GB) are already available via GitHub with instructions for running on typical consumer GPUs, opening up a new playground for neurotech and BCI researchers.