TL;DR:
- Open-source development: The tool allows processing brain-computer interfaces locally on the user’s hardware without sending data to the cloud.
- Competition results: The development team earned fourth place out of 466 participants in the global Brain-to-Text ’25 competition held in February.
- Recorded accuracy: The system achieved a word error rate of 1.78% during the decoding of electrical brain signals.
Tether has just introduced an open-source engine designed to translate neural signals directly into readable text using the QVAC AI ecosystem, which they named Tether BrainWhisperer.
Tether AI's brain-to-text engine is now 100% open-source and available.
This is a larger, more complex version of the same research model placed 4th of 466 teams in an international challenge.
All in QVAC. https://t.co/MAK6Ax6I9z— Paolo Ardoino 🤖 (@paoloardoino) July 16, 2026
The software was developed and managed by Tether EVO, the company’s artificial intelligence research division. According to the ecosystem’s official report, the primary objective is to offer privacy-oriented technological alternatives that directly reduce dependence on centralized cloud service providers.
Local Neural Processing
The tool works by connecting to brain-computer interface (BCI) technologies. Instead of transferring sensitive information to remote servers for processing, the engine performs the entire inference process locally using native QVAC libraries.
The structural design of the software is based on the whisper.cpp optimization, derived from the Whisper transcription model developed by OpenAI. However, engineers adapted the decoding components to specifically interpret electrical fluctuations and signals from the brain rather than traditional human speech. The development team indicated that neural data is among the most sensitive information an individual can generate, which is why keeping it protected locally is projected to prevent the common risks of mass storage within artificial intelligence corporations.
Scientific Validation and Technical Performance
The technical effectiveness of this model was publicly evaluated against hundreds of specialized machine learning teams. Official data from the global Brain-to-Text ’25 competition confirms that the Tether EVO team earned fourth place out of a total of 466 participants in February, after recording a word error rate (WER) of only 1.78% in the interpretation of electrocorticography signals.
Furthermore, the theoretical advances behind the algorithm have just received validation from the international academic community. Engineers from Tether’s scientific division recently published a detailed article in the Journal of Neural Engineering focused on speech decoding from cross-subject data, which, according to current trends, exposes the organization’s interest in pushing these systems beyond traditional laboratory standards.
The deployment of this interface directly complements other lightweight solutions previously launched by the tech firm throughout this year. These prior tools include the QVAC MedPsy medical model and the TurboQuant memory optimizer, aimed at enabling the efficient execution of complex language models on commercial mobile or laptop devices without the need for external supercomputers.
The long-term vision aims to integrate these autonomous local agents with decentralized financial rails like Bitcoin and USDT, allowing for native transactions and secure software operations without intermediaries.
The application’s code repository is already available to the general public on standard development platforms. Independent developers and neuroscience sector researchers can audit the libraries, modify encoder parameters, and build derivative applications from this release, marking a verifiable milestone that will serve as a foundation for upcoming safety evaluations the company will present in the next cycle of decentralized technology conferences.
