QVAC SDK

QVAC SDK Overview

QVAC SDK is the world's first open-source cross-platform framework for running, training, and fine-tuning large language models (LLMs) directly on mobile devices and personal computers. Developed by Tether, QVAC gives developers and researchers a complete toolkit for working with AI without any cloud connection.

Why QVAC Is a Game-Changer

Before QVAC, running — let alone fine-tuning — LLMs on mobile devices was considered virtually impossible. Existing solutions like llama.cpp could run models but not train them. QVAC solves this with its innovative BitNet LoRA technology, enabling fine-tuning on resource-constrained devices — from smartphones to laptops.

Key Features

Local AI Without the Cloud. QVAC lets you work with AI models fully autonomously after the initial download. This means complete data privacy — your queries and data never leave your device. For businesses handling confidential information, this is a critical advantage.

Fine-Tuning on Mobile Devices. BitNet LoRA technology makes model fine-tuning possible on smartphones. You can adapt a model to your tasks right on an iPhone or Android device, without needing powerful GPU servers. The process is optimized for minimal memory and power consumption.

Multimodal Capabilities. QVAC supports not just text generation but also computer vision, OCR (text recognition in images), text-to-speech (TTS), speech-to-text (STT), and machine translation. All modalities run locally on the device.

Cross-Platform Support. The SDK supports all major platforms: iOS, Android, Windows, macOS, and Linux. A unified API lets you write code once and run it anywhere. Native framework integration (Swift, Kotlin, C++, Python) makes adoption straightforward.

P2P Inference. A unique decentralized inference feature lets you combine multiple devices for collaborative work with large models. For example, several phones can pool resources to run a model that wouldn't fit on a single device.

Use Cases

For Mobile Developers. Integrate AI capabilities into iOS and Android apps without cloud costs. Chatbots, autocomplete, image analysis — all working offline.

For Researchers. Rapid prototyping and testing of models on edge devices. Explore on-device AI without expensive cloud infrastructure.

For Business. Deploy AI solutions in environments with limited or no internet connectivity: manufacturing facilities, medical institutions, field operations.

For Privacy Enthusiasts. A fully local AI assistant that never sends data to third parties. Perfect for personal data, medical records, and financial information.

Getting Started

Setting up QVAC SDK takes just a few minutes. Connect the SDK via npm, pip, or CocoaPods depending on your platform, download a model from the built-in catalog, and start working. Documentation at qvac.tether.io includes detailed guides and examples for every platform.

Supported Models

QVAC supports popular open-source models in optimized formats: Llama, Mistral, Gemma, Phi, and others. The model catalog is constantly expanding, and the quantization system lets you run models of various sizes depending on your device's capabilities.

License & Community

The project is distributed under the Apache 2.0 license, allowing free use in commercial projects. The GitHub repository is actively maintained and open to community contributions.