Private dictation for Windows.

Version 0.2.0 · Windows 10/11

Speak. It appears.

Hold Right Ctrl, speak naturally, release. Your words land at the cursor, transcribed by Whisper on this PC, not in a cloud.

Free · MIT licensed · 1.33 GB installer

The Draft window while listening: the status is Listening, an amber timer runs, a live input meter peaks, and the hint reads Release to transcribe.

The Draft window while listening, from the 0.2.0 build.

Audio never leaves this PC · No account · No telemetry · MIT licensed

The privacy model

One key.
Then the audio is gone.

No modes, no wake word, no window to switch to. Hold, speak, release. Draft handles the rest at the cursor.

Hold

Right Ctrl by default. Right Alt, F8, F9, Pause, and Scroll Lock are one click away.

Speak

The input meter is the real level Draft hears. Esc cancels the take.

Release

The finished transcript is pasted wherever you were typing.

Live preview is optional and off by default. Each preview is another full model pass, on the GPU when there is one; the final transcript always uses the complete clip.

The Draft window in its Ready state, showing the Right Ctrl key, a quiet input meter, and the status Ready.

ReadyActual capture

Tap the key, or hold Right Ctrl, to see the live recording state.

Showing the ready state

The whole app is one window.

Dictate and Settings are views inside one fixed window, not separate apps.

Five screens from the shipped app.

Dictation

Draft’s Dictation settings, showing the shortcut presets, paste at cursor, live preview, and start at sign-in.

Six shortcut presets, cursor paste, live preview, and sign-in launch. Live preview stays off by default.

Recognition

Draft’s Recognition settings, showing the bundled Whisper large-v3-turbo model and Vulkan backend.

Whisper large-v3-turbo, F16. Vulkan when the machine supports it, with a CPU fallback when it does not.

Appearance

Draft’s Appearance settings, showing the light, dark, and system theme choices.

Light by default. Dark is a designed palette, not an inversion, and the app can follow Windows instead.

Advanced

Draft’s Advanced settings, showing model load time, last dictation, logs, and the draftctl path.

Model load time, last dictation, logs, and the draftctl path. Nothing hidden behind a second menu.

Credits

Draft’s Credits settings, listing the interface, inference engine, speech model, and typefaces used.

Every project and license in the app, with the installed license texts one click away.

No cloud.No account.No audio kept.

The microphone is open only while the key is held. Whisper runs on this PC, and the recording is discarded once the text is ready.

Draft’s Privacy settings, showing local-only processing, the model path, and the clear history action.

The Privacy tab, captured in the shipped app.

Microphone → local Whisper → cursor

Only text and duration can stay in local history. Never audio, window titles, or full URLs.

The one network path is the model. It is included in the Windows installer; a build from source without it fetches the pinned F16 file once and verifies its SHA-256.

Fast enough to disappear.

The model stays warm on the GPU. A three-second clip is usually finished before you look up: 239 ms at p95 in the release benchmark.

Three-second clip, p95

239ms

4.17× faster than the previous build on the same clip.

Two warm-up passes, then five measured runs per clip for Draft 0.2.0: 231 ms median, 239 ms p95. The v3.5 baseline used ten warm runs. Decode only, before the paste; a 60‑second clip finishes in 1.19 s p95. “v3.5” is the previous app, same model, decoded on CUDA rather than Vulkan.

Draft 0.2.0Previous app (v3.5)p95, decode only

3 sclip

4.17×
3 s clip: Draft 239 milliseconds p95, the previous app 996 milliseconds p95, 4.17× faster.

5 sclip

3.97×
5 s clip: Draft 234 milliseconds p95, the previous app 929 milliseconds p95, 3.97× faster.

15 sclip

3.04×
15 s clip: Draft 375 milliseconds p95, the previous app 1139 milliseconds p95, 3.04× faster.

30 sclip

2.95×
30 s clip: Draft 543 milliseconds p95, the previous app 1601 milliseconds p95, 2.95× faster.

60 sclip

2.56×
60 s clip: Draft 1186 milliseconds p95, the previous app 3036 milliseconds p95, 2.56× faster.

Intel Core i7-13620H, GeForce RTX 4060 Laptop 8 GB, Windows 11, Vulkan, F16, warm model. Your hardware will differ. Method and raw data.

Draft app icon

Private dictation. On your machine.

The installer includes the speech model, so the first dictation needs no setup and no network.

Windows 10/11 · x64 · 1.33 GB · Free & MIT

Unsigned installer · SmartScreen may warn · verify first

Draft is MIT licensed and written in Rust. The interface runs on GPUI, inference on transcribe.cpp, and the words come from Whisper large-v3-turbo. Read the source.

Version
0.2.0
Released
October 21, 2026
File
Draft-Setup-0.2.0.exe
Size
1,425,663,461 bytes
License
MIT
Code signing
Not signed yet

Draft 0.2.0 is an early Windows x64 build, and the installer is not code-signed. Windows may show a SmartScreen warning. Verify the SHA-256 first.