QVAC vs OpenAI

These two are often mentioned in the same breath, but they solve different problems. OpenAI gives you access to some of the most capable models in the world — over a network, from a data center. QVAC gives you AI that runs entirely on the user's device, with no network round trip at all. Neither is "better" in the abstract; the right one depends on what you're building.

Here's a fair look at both.

The one-line difference

OpenAI is a cloud API: you send a request, powerful models in OpenAI's data centers process it, and the response comes back. QVAC is an on-device SDK: models run locally on the phone, laptop, or server your app is installed on, and nothing is sent anywhere.

Side by side

	QVAC	OpenAI API
Where inference runs	On the user's device	In OpenAI's cloud
Data leaves the device?	No	Yes (sent to the API)
Works offline?	Yes, after model download	No
Per-request cost	None (runs on user hardware)	Metered per token
Model capability ceiling	Bounded by device hardware	Very high (frontier models)
License	Open-source (Apache 2.0)	Proprietary service
Multi-modal	Text, embeddings, transcription, OCR, translation, more	Text, vision, audio, image gen
Platforms	iOS, Android, macOS, Windows, Linux	Any client that can call an API

Where OpenAI is the stronger choice

If your app needs the absolute frontier of reasoning or knowledge — the largest, most capable models — a cloud API delivers a ceiling no phone can match today. Hardware limits on-device model size, so a small local model will not match GPT-class frontier performance on the hardest tasks. If you're building something where raw model capability is the product and users expect an always-connected experience, cloud makes sense.

You also offload all the compute. Your users don't need capable hardware; OpenAI's servers do the work. The trade is that you pay per request and your users' data travels to a third party.

Where QVAC is the stronger choice

If the data is sensitive — notes, health, finances, anything personal — the cloud model means that data leaves the device. QVAC removes that entirely: there's no server for the data to reach, so there's nothing to leak, log, or subpoena. Privacy becomes structural rather than a policy promise.

QVAC also has no per-request cost, because inference runs on the user's own hardware. For an app with heavy AI usage, cloud token costs scale with every user and every action; on-device inference doesn't. And QVAC works offline — once a model is downloaded, the AI keeps working with no connection, which cloud APIs can't do by definition.

The honest trade-off

You're choosing between capability-in-the-cloud and privacy-and-independence-on-device. OpenAI gives you a higher capability ceiling at the cost of data leaving the device and metered pricing. QVAC gives you privacy, offline function, and zero marginal cost at the cost of being bounded by device hardware.

Many products will use both: cloud for heavy, non-sensitive tasks; on-device for anything personal. They're not mutually exclusive.

Our upcoming apps — private notes and private finance — will handle exactly the kind of data that shouldn't leave your device. For us the choice was clear: the value isn't a frontier model answering trivia, it's a trustworthy model working over your most personal information without that information ever moving. QVAC lets us make privacy the architecture, not a setting.

If you're building something with sensitive data, here's how QVAC works and what we're building on it.