QVAC is Tether’s open-source local AI SDK for running artificial intelligence entirely on-device, with no cloud and no API keys. It packs LLM inference, retrieval, media generation and speech into one cross-platform toolkit. The project has about 703 stars on GitHub and is under active development in 2026.
Built for a decentralized, privacy-first approach to AI, QVAC lets developers embed capable models directly into applications. It runs GGUF-format models for text inference, supports RAG over local data, speech-to-text based on Whisper-style models, and image, music and video generation, and it can even fine-tune models locally. A distinctive feature is peer-to-peer inference, where devices can share compute without a central server. SDKs are provided for Node.js, Python and TypeScript, and the runtime targets Linux, macOS, Windows, Android and iOS, making it one of the broadest on-device AI toolkits available from a major crypto-industry engineering team.
Tether develops QVAC within its broader push into decentralized, self-sovereign infrastructure, applying the same principle it follows in digital finance: users should hold their own keys, their own data and, now, their own models.
QVAC is free and open source under the Apache-2.0 license. Since inference happens on your own devices, there are no per-token charges or subscription fees at all.
QVAC is for developers building AI features that must work offline, protect user privacy or avoid cloud API costs: mobile apps, edge devices and decentralized applications. It is also a good fit for teams in regions with unreliable connectivity or strict data rules.