Is QVAC Production-Ready? An Honest Review

We're building Local Notes and Local Money on QVAC — both coming soon. So we have an obvious interest in you thinking well of it.

Which is exactly why this needs to be an honest assessment rather than a sales pitch. If you're evaluating QVAC for something real, here's what we'd want to know in your position.

The short answer

Yes, with caveats, for the right kind of project.

QVAC is pre-1.0 software from a well-resourced organization, moving very fast, with real capability and thin community infrastructure. That combination has specific implications depending on what you're building.

The signals, read honestly

At the time of writing, the repo shows:

308 stars, 78 forks — modest. This is not a project with mass adoption.
1,927 commits, 195 releases — very active. Releases are frequent, sometimes daily.
13 open issues, ~100 open pull requests — the PR count is the interesting number.
Only 2 watchers — almost nobody is tracking this closely.
Apache 2.0, with CONTRIBUTING.md and SECURITY.md present.
~33% TypeScript, ~32% JavaScript, ~26% C++ — the C++ share tells you where the real work is.

How to read those numbers

High commits + high releases + low stars means a serious team building fast, before the world has noticed. That's different from an abandoned project with low stars, and different again from a hyped project with high stars and no substance.

~100 open PRs against 13 issues is unusual. It suggests most work is happening internally — the team is opening PRs against their own monorepo — rather than a backlog of unreviewed community contributions. It's a sign of velocity, not of neglect.

2 watchers is the number that should surprise you. Almost nobody is paying close attention to a Tether-backed open-source AI SDK with brain-computer-interface and robotics support. Make of that what you will.

What's genuinely solid

The core inference paths. Text generation, embeddings, RAG, transcription, and OCR are the well-trodden roads. We've shipped on them. They work.

The architecture. The Bare worker design is coherent, and it pays off in ways that aren't obvious until you see delegated inference work — the RPC boundary means "run on a peer" is a change of address, not a rewrite. That's good design, not luck.

Cross-platform actually delivers. The same code really does run on iOS, Android, macOS, Windows, and Linux. This is the main reason to choose QVAC, and it holds up.

The production plumbing exists. Cancellation, runtime lifecycle, profiling, logging, diagnostics, resumable sharded downloads. Pre-1.0 projects usually don't have these. Their presence signals a team that has shipped mobile software before.

The docs are good. Genuinely. Better than most 1.0 projects.

Tether ships on it themselves. QVAC Workbench and QVAC Health are real applications built on this SDK. Dogfooding at that scale is meaningful evidence.

What isn't ready

The API changes. 195 releases pre-1.0 means signatures move. Pin your versions. Read release notes before upgrading. Budget for migration work. This is the single biggest practical cost of adopting QVAC today.

Frontier capabilities are frontier. BCI, VLA, and video generation are documented and shipped, but they are not the well-tested paths. Treat them as capable foundations for research and prototyping, not as production-hardened features.

Documentation has inconsistencies. The sidebar nav omits BCI, VLA, video generation, and image classification on some pages while including them on others. The README's "complete user docs" tip links to a path that doesn't match the live docs. Small things, but they tell you the docs are moving faster than they're being proofread.

The community is tiny. 2 watchers. When you hit an obscure bug, there is no Stack Overflow answer waiting for you. You will be the first person to hit it. This is the real cost, and it's larger than the API churn.

Third-party ecosystem is nearly nonexistent. Few plugins, few tutorials, few integrations. You're early.

Mobile edge cases will find you. Memory pressure, background/foreground transitions, model lifecycle on low-end devices. We've found bugs here. If you ship a real consumer app, you will too.

Who should build on QVAC today

Yes, if:

Cross-platform on-device AI is a requirement, not a preference. Nothing else does all five platforms from one codebase.
Privacy is a core product commitment, not a feature bullet.
You need multiple modalities (text + speech + vision) and don't want to integrate four separate mobile ML libraries.
You want delegated inference — genuinely nobody else offers this.
You can absorb some API churn and debug things yourself.
Being early is strategically valuable to you.

No, if:

You need maximum model capability and your users are always online. Use a frontier cloud API.
You need in-browser inference. QVAC isn't a WASM library.
You need a large community and Stack Overflow answers today.
You're building something where a breaking change next month would be catastrophic and you can't pin and migrate.
You only need desktop local LLM inference and nothing else. Simpler tools exist for that narrow job.

What we actually do

We pin versions. We read every release note. We test on low-end devices before anything else. We file bugs when we find them — which is often, because we're doing things nobody else is doing on this SDK yet. We keep our own integration layer thin enough that an API change costs us hours, not weeks.

And we accept that being early means being the person who finds the bug. That's a real cost, and for us it's worth paying, because the alternative — uploading our users' notes and bank statements to a third party — was never acceptable, and no other SDK does what this one does across five platforms.

The trajectory

The honest way to think about this: QVAC is not a project that will quietly die. It's backed by an organization with substantial resources and an explicit strategic commitment to it, it ships constantly, and Tether builds their own consumer apps on top of it.

The question isn't whether QVAC will mature. It's whether you want to be building while it does, with the costs and the advantages that carries.

We decided we did. Whether you should depends on which of the lists above you're on.

Verify this yourself

Everything above is from public sources on the day we wrote it. Numbers change. Check the repository, read the release notes, and join the Discord before you commit.

Don't take our word for it. We have an interest.

See also How QVAC Works Under the Hood, QVAC Troubleshooting, and Inside the QVAC Monorepo.