Everything in one app

Your voice never leaves your computer

That sentence is the whole pitch, and it holds. Three things reach the network, and only when: a licence check at launch, an update check at launch, and the model downloads you start yourself. No audio, no text, ever.

• No audio. No text. Ever.
• Once the models are on your disk and the update check is off, nothing goes out at all.
• No network, no problem.

Runs on the laptop you already have

No workstation required. The small model runs on any 4 GB graphics card, and the whole thing runs with no graphics card at all — the floor is 8 GB of RAM, a four-core x86_64 processor, and 5 GB of disk.

• NVIDIA — Vulkan 1.2+ GPUs
• AMD — Vulkan 1.2+ GPUs
• On Linux it ships as a single portable file with nothing to install.
• On Windows it is a normal installer that needs no administrator rights.

Thirty languages, twenty-two Chinese dialects, on your machine

It transcribes in thirty languages and recognises twenty-two Chinese regional dialects on its own, without being told which one is speaking.

All of it runs on your machine, with no per-minute fee and no upload.

Translation in 38 languages — new in v2.0.0

Translation in thirty-eight languages — new in v2.0.0

• Speak Polish, get English. Speak English, get Japanese.
• 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, as many target languages as you want.
• The translation holds up against the cloud giants. Powered by Tencent's Hunyuan MT2, at quality size it reaches 97.9% of Google Gemini 3.1 Pro on the FLORES-200 benchmark, and beats it on real-world and minority-language translation.
• Subtitles, too — plain text or SRT, timed by sentence, or word by word when you need to edit precisely.
• And translating a subtitle file keeps every timing intact.

Your machine, your call

Two model sizes, a choice rather than a tier: the small one for laptops and integrated graphics, the larger one for accented or noisy audio. You can switch at any time. And transcription and translation each pick their own processor, so one can run on the graphics card while the other sits on the CPU.

• Small — for laptops and integrated graphics
• Large — for accented or noisy audio

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.

Noise reduction

Noise reduction is there for recordings made in a noisy room.

• It is off by default, because it was measured to make short, clean microphone clips worse.
• Turn it on when the room is loud.
• Leave it off when your recording is fine.
• Off by default — toggle per-recording or always-on

It learns from the corrections you already make

Most voice apps ask you to read sample sentences. This one trains on the fixes you already make: correct a misheard word and that correction becomes the example.

• No sample recording, no homework.
• Training is one click
• Runs on your own hardware
• The new model is ready to switch to when it finishes
• No recording sessions
• No "please repeat the following paragraph"
• Just your voice, and the fixes you were going to make anyway

The keyboard that listens

Type wherever the cursor is — the editor, the terminal, a chat box, a form, any field that takes a keyboard. And if you want, it types the translation instead of the transcription: say it in Polish, and the English lands in the page.

• The microphone
• An audio or video file
• The audio playing through your speakers
• Works with any text field, editor, terminal, or chat
• X11 and Wayland on Linux, native input on Windows

Your words, spelled your way

Brand names, jargon, colleagues' names. Pin them once and they stop coming out wrong. The same list steers the transcription and the translation, so a name is never localised by accident. Correct a misheard word once and it is training data — you are never asked to read it out for practice.

• Add terms in Settings — one per line
• Improves recognition of proper nouns and abbreviations
• Preserves brand names and technical terms in translations
• No retraining needed — applied at inference time

The agent gets ears

Give your AI agent the ability to listen. Through MCP it can transcribe, translate and record a meeting of any length, on your machine, without an API key.

• Transcribe audio/video files, mic recordings, or system audio
• Translate text, SRT, or VTT (bilingual mode supported)
• Switch ASR or MT compute device on the fly
• List and switch personal voice profiles
• Read/write any app setting from the agent

A three-hour call is no different from a three-minute one — nothing to babysit, and if you forget it was recording, nothing is lost: it finishes by itself and keeps the transcript for you.