What Is QVAC — and Why We Built localhost/AI On It

Most AI apps send your data to a server. You type, it uploads, a data center answers, the answer comes back. That round trip is where your privacy goes.

QVAC is built to remove the round trip entirely. And it's the engine underneath everything we make at localhost/AI.

This page explains what QVAC is, how it works, and why we bet our whole studio on it — in plain language, whether you're a developer evaluating the SDK or someone deciding whether to trust an app with your notes and your money. Prefer an even gentler start? Try the beginner's guide first.

What QVAC is, in one sentence

QVAC is Tether's open-source SDK for running AI directly on your device — language models, transcription, OCR, embeddings, and retrieval — with no server, no account, and no data leaving the hardware in your hand.

That's the whole idea. The model comes to your data instead of your data going to the model.

Who makes QVAC — and who makes the apps

Three names get used around our apps, and they're three different things. Keeping them straight matters:

• QVAC is the engine. It's built by Tether (the company behind USD₮), released open-source under Apache 2.0, and it's the technology that runs AI on the device. It is not ours — we build on it.
• localhost/AI is the studio. That's us: the brand, the design, the apps, the experience. We build consumer software on top of QVAC.
• LOCALHOST 41 INC. is the legal entity behind localhost/AI.

So: QVAC is Tether's. Local Notes and Local Money are ours (both coming soon). We're a studio building on QVAC the same way a game studio builds on a graphics engine it didn't write.

How QVAC actually works

Under the hood, QVAC is a unified toolkit over several best-in-class local inference engines — text generation, embeddings, speech-to-text, OCR, translation, and retrieval-augmented generation (RAG) — all exposed through one interface, running from a single codebase across iOS, Android, macOS, Windows, and Linux.

A few properties matter for anyone deciding whether to trust it:

Nothing leaves the device. All processing happens locally. Your prompts, documents, and images are never sent to an external server. The only network activity is the initial model download — and after that, the app works fully offline. You can turn on airplane mode and the AI still answers.

No account, no login. Because there's no server, there's nothing to sign up for. You download the app, it fetches a model on first run, and you're working. There's no user database, because there is no user database anywhere.

It's open-source. QVAC is public and auditable under Apache 2.0. You don't have to take a privacy promise on faith — the engine is inspectable.

Privacy is architecture, not a setting. This is the part we care about most. In most apps, "privacy" is a toggle you flip to restrict what a server does with data it already received. With QVAC, there's no server for the data to reach. There's nothing to leak because there's nowhere for it to leak to.

Why we built localhost/AI on QVAC

We started localhost/AI on a simple conviction: the most personal software you own — your private notes, your financial life — should not live in someone else's cloud.

For years, that conviction meant a compromise. You could have privacy or you could have good AI features, but the good AI features needed a server, and the server needed your data. QVAC is the first foundation we found that let us refuse the compromise. We can build real on-device intelligence — chat over your own notes, receipt scanning, expense categorization, voice logging — with the model running on your phone and your data never moving. So we're building our apps to mirror that architecture end to end, including the no-login, download-a-model-on-first-run experience that QVAC's own apps use. The engine's promise and our product promise are the same promise.

What we've built on QVAC

Local Notes (coming soon) — a private notes and workspace app, deliberately simpler than Notion. Notebooks, notes, a focused set of blocks, tags, and search. The headline feature is AI chat over your own notes — across everything, one notebook, or a single note — plus inline writing actions like improve, summarize, fix, and continue. All on-device. Write like no one's watching.

Local Money (coming soon) — private personal finance in the spirit of Copilot Money, but with no bank linking, ever. A net-worth dashboard, transactions, and on-device AI for categorization, receipt OCR, and voice-logged expenses, plus an "ask your money" chat. Your data gets in by importing a statement file, importing an export from another app, or entering it yourself — never through a live bank connection, which would break the on-device promise.

A third app in the same family is coming soon.

localhost/AI is a local-first AI studio. We're building consumer apps where the AI will run entirely on your device. Learn more about Local Notes and Local Money, both coming soon.