For the past several months, we have been building toward a simple idea with very large implications: AI should run on your terms, on your hardware, under your control. Today, we are opening a major new chapter in that effort with the launch of the QVAC SDK, the developer entry point into QVAC, Tether’s local-first, peer-to-peer AI ecosystem. Developers can start there and go deeper through the full QVAC documentation.
This launch is not about adding one more AI framework to an already crowded market. It is about making a different future possible. A future where intelligence does not have to live behind someone else’s API, inside someone else’s data center, or within infrastructure you do not control.
Privacy matters, of course. But for us, privacy is a byproduct, not the main objective. The deeper goal is freedom. If your intelligence depends on centralized infrastructure you do not control, then it is not really yours. In the same way Bitcoin made self-custody impossible to ignore, QVAC is built around the idea of self-custody for intelligence.
That belief shaped every layer of the product. QVAC is designed so developers can build local and peer-to-peer AI applications without carrying the cognitive burden of platform-specific choices, fragmented runtimes, or a maze of incompatible engines. The SDK provides a unified JavaScript and TypeScript interface for running AI across desktop and mobile environments, including Node.js, Pear, and Expo. A big reason we can be cross-platform from the start and expose one unified API is the Pear Runtime, built by Holepunch. You write against one consistent surface and let QVAC handle the complexity underneath. And this is only the beginning. We are already planning SDKs for Swift, Kotlin, Python, and other languages, so the same philosophy can extend well beyond the JavaScript ecosystem.
That developer experience is not an afterthought. It is the core of the value proposition. Vision alone is not enough. If local AI is going to matter, it has to be easier to build with, easier to ship, and easier to scale into real products. That is why the QVAC SDK brings together capabilities such as completion, text embeddings, translation, transcription, text-to-speech, OCR, multimodal interactions, RAG, and delegated inference into one coherent package. Just as importantly, every capability in the SDK is built from the ground up as a plugin. That means developers can extend the SDK’s behavior themselves, whether that means introducing a new engine, supporting a new model family, or adding entirely new capabilities.
It is also one of the reasons we invested in QVAC Fabric, our inference engine forked from llama.cpp and evolved to support a much broader set of use cases in a more unified way. Rather than forcing developers to constantly juggle different engines, formats, and platform-specific integrations, Fabric gives us a foundation we can keep extending. The result is a cleaner developer experience today and a more ambitious roadmap tomorrow.
We wanted to validate early that we were moving in the right direction, so a few months ago we ran a fully remote hackathon around QVAC. In just 48 hours, around 40 participants built an impressive set of projects that pushed well beyond toy demos. Teams created ideas such as AI-assisted vision experiences for visually impaired users and natural-language home automation systems capable of controlling an entire environment. That was an important signal for us. It showed that developers were not just interested in the philosophy. They were ready to build serious applications with this stack.
At the same time, we have continued pushing the underlying research forward. Our work is not limited to wrapping existing tools in a nicer SDK. We are actively advancing the state of the art in edge inference and fine-tuning. A recent example is our BitNet support for Fabric, which was met with strong interest and reflects our broader commitment to making efficient, high-performance local AI available on real-world hardware, especially consumer devices. That matters because the point is not just technical progress. The point is democratizing access to local AI.
This combination of product and research is what makes QVAC different. We are trying to solve two problems at once: giving developers a better way to build, and expanding what is technically possible at the edge. We believe those two efforts have to move together. Better infrastructure without usability stays niche. Better usability without deep technical progress eventually hits a wall.
The ecosystem is already taking shape. QVAC Workbench and QVAC Health have been available for some time and demonstrate what this approach looks like in practice.
QVAC Workbench pioneered our concept of delegated inference: one device can provide inference services while another connects as a thin client in a fully peer-to-peer way, and that capability is baked directly into the SDK. It has been available in Workbench since the first launch in October 2025. We have also already tested crypto payments in practice as part of that vision by integrating WDK into a coffee demo at the Plan B conference in El Salvador, showing how local AI, peer-to-peer infrastructure, and payments can come together in a single experience. We also showed that demo in our talk here: QVAC – Sovereign AI for the Universe.
QVAC Health shows what it means to keep deeply personal data on-device while still making it useful through natural language, OCR, charts, and personalized insights. It can also connect to wearables directly over BLE for privacy reasons, so data does not need to pass through the cloud before becoming useful. These are not theoretical examples. They are proof that the model works.
The launch of the QVAC SDK is an invitation to developers who want more than access to AI. It is for those who want ownership, portability, resilience, and the ability to build without asking centralized infrastructure for permission. If the next era of software is going to be shaped by intelligence, then that intelligence needs to be open, local-first, and sovereign by design.
That is what QVAC is about. Not rented intelligence, but owned intelligence. Not dependence, but freedom. And starting today, any developer in the world can build with it.
