Published models: free for commercial use. Custom models: built for you.
VoxRT is freemium: everything published in our GitHub org is free to use — including in commercial production apps — with no per-user, per-device, or per-detection fees. Custom models tuned to your product are a paid engagement.
What's free, what's a paid engagement
Free, today, for commercial production use: the wake-word SDKs with the reference "Hey Assistant" model (iOS, Android, Linux, browser), the Silero-v5 VAD SDKs (iOS, Android), the streaming ASR SDKs (iOS, Android, English), and LinnetTTS text-to-speech with its built-in voices (Android, iOS, Linux, Python). No license key, no account, no usage metering — install from Swift Package Manager, JitPack, PyPI or npm and ship.
Paid engagements: models built for your product — a custom wake phrase ("Hey YourBrand"), a keyword-spotting vocabulary, a speech-to-intent model on your intents, a custom text-to-speech voice, additional languages, and OEM or bulk-device deployments. Tell us what you're building via the Try VoxRT form and we'll scope it.
Three layers, one simple rule
Every VoxRT SDK has the same three-layer structure. The rule: if it's published, it's free to use commercially.
| Layer | What it is | License |
|---|---|---|
| SDK wrappers | The source you integrate — Swift, Kotlin, Python, Node.js, Go, C/C++ headers and examples. | Apache-2.0 |
| Runtime | The compiled VoxRT inference runtime (xcframework / .so / .wasm). Proprietary; the unmodified runtime may be incorporated and redistributed in your applications and products. | Free for commercial use no per-user fees |
| Model weights | Varies by product: reference wake word (VoxRT in-house, free to ship with the SDK), Silero VAD v5 (MIT), ASR FastConformer (CC-BY-4.0, NVIDIA attribution), LinnetTTS (Linnet Weights License). | Per product |
Licensing, answered
Are VoxRT published models free for commercial use?
Yes. The published SDKs are free to use in commercial production apps: the SDK wrapper source is open under Apache-2.0, and the compiled VoxRT runtime is free for commercial use with no per-user, per-device, or per-detection fees. When you need more, custom models — wake phrases, keyword vocabularies, speech-to-intent, additional languages — are paid engagements.
What do I pay for?
Custom models: wake phrases, keyword vocabularies, speech-to-intent models, text-to-speech voices, additional languages, and OEM or bulk-device deployments. Everything published in the VoxRT GitHub org is free to use, including commercially.
What licenses cover the model weights?
It varies by product: the reference wake-word model is VoxRT's own proprietary model (free to ship inside your applications, no per-installation fees), the Silero VAD v5 weights are MIT, and the ASR weights are CC-BY-4.0 — derived from NVIDIA NeMo, attribution required. The LinnetTTS weights are under the Linnet Weights License.
How is LinnetTTS licensed?
Its code, including the runtime, is Apache-2.0. Its weights, tokenizer, and voices are under the Linnet Weights License, which is not an open-source license. It's free to use the unmodified weights with the built-in voices inside your own products, including commercially. Custom voices, modified weights, training other models, and offering general-purpose text-to-speech to others need a separate license. The license text governs; see the product page for a summary.
Do I need a license key or account to ship?
No. The SDKs are distributed publicly (Swift Package Manager, JitPack/Gradle, PyPI, npm) and the models via GitHub Releases. There is no per-user metering, license server, or usage reporting.