You're talking. The words are landing — in your editor, your terminal, a chat box, a form, wherever the cursor sits. And every one of them stays on the machine you said them on.
Brethof Voice Pro is voice-to-text and translation that runs entirely on your own computer. That isn't a privacy feature. It's the whole product.
One program, the work of four. It transcribes your voice, translates it across 38 languages, subtitles your videos, and types for you wherever a keyboard would. It runs offline — on a plane, in a hotel with bad wifi, through a tunnel — and on the laptop you already own, with or without a graphics card. No subscription. No per-minute meter. No uploads. Give your AI agent ears, too: it can transcribe, translate, and record a meeting of any length through MCP, on your machine, with no API key. Buy it once, and it keeps working.
Your voice never leaves your computer. That sentence is precise, and it's the pitch. Three things reach the network at launch, and only three: a licence check, an update check, and the model downloads you start yourself. The licence check sends a licence key, the update check a version string — no audio, no text, ever. Turn the update check off, and once the models are on disk, nothing goes out at all.
Transcription in 30+ languages, and 22 Chinese regional dialects understood on their own — no setup, no training. And your accent stops coming out wrong. Brethof Voice Pro learns your voice, your jargon, and your proper nouns, on your own machine. Speak your grandmother's Sichuan and it comes out right.
Speak Polish, read English. Speak English, read Japanese. Or say it once and have it typed in English, Japanese, French and Spanish at the same time — one per line, or side by side in the same line. Pick as many target languages as you want; 38 are there, and 23 work in both directions. The translation happens on your machine too, in Tencent's Hunyuan MT2. At the quality size it reaches 97.9% of Google's Gemini 3.1 Pro on the FLORES-200 benchmark, and beats it on real-world and minority-language translation. Two sizes, both optional downloads: a fast one for everyday, a quality one for the words that matter. Subtitles, too. Plain text or SRT, timed by sentence, or word by word when you need to edit precisely. Translate a subtitle file and every timing survives. Subtitle your own video in 38 languages without paying by the minute.
That's why it works where nothing else will. A plane. A hotel with bad wifi. A train through a tunnel. There's nowhere for your recording to go, so it doesn't need to go anywhere. Drop in a two-hour meeting recording and go make coffee; the transcript is there when you're back. Forget you left it recording and nothing is lost — it finishes the job on its own and keeps the transcript.
Two transcription models, and it's a choice, not a tier — you can switch any time.
Most voice software makes you read out sample sentences at setup. This one doesn't. It learns from the corrections you already make: you fix a misheard word, and that fix becomes the training example. Pin your brand names, your jargon, your colleagues' names once, and they stop coming out wrong — in the transcript and in the translation, because the same list steers both. A name is never localised by accident. Training is one click, it runs on your hardware, and the new model is ready to switch to when it's done.
Buy it once and it's yours. A perpetual licence with a year of updates included.
A noisy room stops being a problem. Noise reduction is built in, for recordings made in a café, a trade show floor, a car. It's off by default, because we measured it: on short, clean microphone clips it makes things worse. For the two-hour meeting with the espresso machine behind you, switch it on.
The voice keyboard types wherever your cursor is — a browser, an IDE, a chat app, any field that takes a keyboard. And it can type the translation instead of the transcription. Say the sentence in your language, and the right words land in the form, in the language the form needs. One line, no cutting and pasting.
The 0.6B is the everyday one. It runs on any 4 GB graphics card, on integrated graphics, and on laptops with no graphics card at all. Fast, light, there when you need it.
The 1.7B is the careful one — for accented audio, noisy audio, the two-hour meeting where the words matter. It wants about 6 GB of VRAM, or 8 GB if the quality translation model is loaded alongside.
No GPU at all? It still runs. CPU-only needs 8 GB of RAM, a four-core processor, and about 5 GB of free disk; everything else is 7–9 GB. Any Vulkan 1.2+ card works — NVIDIA, AMD, Intel Arc. And transcription and translation each pick their own processor, so one can run on the graphics card while the other runs on the CPU.
Optional add-ons download on demand from Settings → Models:
Forced Aligner (~540 MB) for word-level timestamps · Hunyuan MT2 Fast (~1 GB) or Quality (~4.3 GB) for translation.
| Feature | Brethof Voice Pro | Dragon | Google STT | Otter.ai | Whisper (OSS) |
|---|---|---|---|---|---|
| 100% local processing | ✓ | ✓ | ✗ | ✗ | ✓ |
| Perpetual license | ✓ | ~ | ✗ | ✗ | ✓ |
| Native Linux support | ✓ | ✗ | ~ | ✗ | ✓ |
| Native Windows support | ✓ | ✓ | ~ | ✗ | ~ |
| 30 ASR languages + auto-detect | ✓ | ✗ | ✓ | ~ | ✓ |
| Offline translation (38 languages) | ✓ | ✗ | ✗ | ✗ | ✗ |
| GPU acceleration (NVIDIA + AMD + Intel) | ✓ | ✗ | N/A | N/A | ~ |
| Personal model fine-tuning (LoRA) | ✓ | ✓ | ✗ | ✗ | ✗ |
| MCP server for AI agents | ✓ | ✗ | ✗ | ✗ | ✗ |
| Built-in noise reduction | ✓ | ✓ | ✓ | ✓ | ✗ |
| Direct text injection | ✓ | ✓ | ✗ | ✗ | ✗ |
| Polished desktop GUI | ✓ | ✓ | ✗ | ✓ | ✗ |
| Typical cost | $49 once | $350+/yr | $17/mo | $17/mo | Free |
No monthly fees. No usage limits. Perpetual license with 1 year of updates included.
No credit card required. Just an email to verify your trial.
Perpetual license. 2 personal devices. 1 year of updates included.
Prices excl. tax. Then $20/year for updates (optional)
Per-seat perpetual license. Team & organization use. 1 year of updates.
Prices excl. tax. Then $20/seat/year for updates (optional)
No. Brethof Voice Pro processes everything locally on your device. No audio or text data ever leaves your computer. There is no cloud component, no telemetry, and no analytics.
Any modern GPU works. NVIDIA, AMD, and Intel Arc all use Vulkan acceleration. You can also run on CPU only, though transcription will be slower. The 0.6B model runs comfortably on integrated graphics or any 4 GB+ Vulkan card.
Start with the 0.6B model — it is the recommended default and runs great on most GPUs (and even on CPU on most modern machines). If you need higher accuracy on accented or noisy audio, switch to the 1.7B model (needs 6 GB+ VRAM). You can switch sizes at any time from Settings → Models without re-downloading.
Yes. Brethof Voice Pro supports both Linux and Windows natively. On Linux it works with X11 and Wayland. On Windows it runs as a standard desktop application.
Your license is perpetual — the app keeps working forever with whatever version you have. The optional $20/year Update Pass gives you access to new features and model improvements. Without it, you simply stay on your current version.
Yes — personal voice training is included in v2.0.0 and runs end-to-end on your machine. Every time you correct a misrecognised word, the {clip, correction} pair is auto-saved to your local training dataset. The main window's training card shows total samples and minutes captured at a glance — click "Start training" in the Training tab to fine-tune a LoRA on your accent. The result auto-exports to GGUF and you switch to it in one click. Free for every paid licence, your voice data never leaves your machine.
14-day free trial. No credit card. No cloud. No compromises.
Local speech-to-text that learns your voice. Perpetual licence. Our flagship.
PAID · flagship
Memory for your AI agents — records, rules, the full history and the graph of decisions, already there when a session starts. It answers before you ask.
PAID · free tier
Print-ready digital models. GLB and OBJ included. Lifetime access.
PAID · digital catalog
Five printers and real capacity. No self-serve shop yet — tell us what you need and we quote it by email.
BY ARRANGEMENT · email us
Our YouTube channel. A cyber-tiger host walks through local AI tools and what they actually do.
CHANNEL · live
Curated GitHub lists for AI coding agents, MCP servers, local AI and Linux for AI. Every entry carries a source link.
FREE · curated
Long-form how-tos for local AI on Linux, Windows and macOS, with the configuration files included.
FREE · live
Negative-curation: practices and tools that waste your time, ranked. Receipts required.
FREE · live
Who we are, why we build privacy-first AI, and what we won't do.