01
Hold the hotkey
Anywhere on your desktop. Default is Alt+Space, change it in Settings.
SpacialVoice 0.1 is live — download for your platform today.
SpacialVoiceTalk it into the
cursor.
SpacialVoice is the SpacialMind dictation app. Hold a hotkey, say the line, release — the transcript appears in whatever window already had focus. Editor, terminal, Slack, browser. No window switch, no copy-paste, no flow break.
Two engines under the same key. Local Whisper on-device for when the audio must not leave the machine. Hosted Whisper for 99+ languages and faster turnaround. You pick per session.
Windows 10+macOS 12+Linux x11/Wayland
SpacialVoice listening
$ git commit -m ""
latency
0.6s
streamed bytes
342 KB
language
en (auto)
01 The loop
Press, speak, release.
One global hotkey, available everywhere. The audio stream stays warm so capture begins in the first frame — no "ready?" pause.
02
Speak naturally
A persistent stream is already open, so capture starts in the first audio frame.
03
Release
SpacialVoice transcribes, applies your dictionary, and types the result at the cursor.
04
Keep going
Every transcript is logged locally with words and WPM. Recopy anything with one click.
02 Engines
Two engines, one keybind.
Privacy when you need it, multilingual reach when you don't. The switch is one click — your hotkey, history, and dictionary stay the same.
on-device
Local Whisper
Six ggml models, from a 75 MB Tiny to a 3.1 GB Large-v3. Runs through whisper.cpp inside the Tauri shell. No network calls, no telemetry, no audio leaving your machine.
offlineenglishprivate
cloud
Hosted Whisper
Streamed through the SpacialMind API to a hosted Whisper Large-v3 cluster. Auto-detect across 99+ languages with sub-second end-to-end latency. Included with Pro.
99+ languagesauto-detect<1s round-trip
03 Surfaces
Anywhere the cursor blinks.
SpacialVoice doesn't require integration. It uses the same OS-level keyboard path as your real keyboard, so it works in the app you already use.
Editors
VS Code, Cursor, Zed, JetBrains, Vim — any text input surface.
Terminals
Inject commands, commit messages, and shell snippets straight into your shell.
Chat & docs
Slack, Discord, Notion, Gmail, Linear — talk through the long ones.
Browsers
Forms, GitHub PR descriptions, comments. If the cursor blinks, it works.
04 Dictionary
Teach it your jargon.
Define replacement rules for the things Whisper can't know. SpacialVoice applies them after transcription, before injection — so the typed text matches how you actually write code.
You sayIt types
"use effect"useEffect
"spacial mind"SpacialMind
"tauri two"Tauri v2
"dot env".env
"new line"\n
"wasm"WASM
05 Built for it
Quiet, fast, private.
100%
private in local mode
6
Whisper models
<10ms
capture cold-start
3 OS
Win / mac / Linux
Tauri 2 + Rust core
One Rust crate per concern: capture, engine, hotkey, injection. Same code on all three OSes; native bindings only where required.
whisper.cpp under the hood
Local mode uses the same hyper-optimized engine ggerganov ships, wired in through whisper-rs. Hardware acceleration where the device has it.
Nothing on disk in cleartext
API keys live in the OS keychain. Audio is held only in RAM until transcription finishes. History is local SQLite, on the disk you already trust.
06 Questions
Frequently asked.
What is SpacialVoice?+
A desktop dictation app for builders. Hold a hotkey, talk, and the transcript types itself into the app you're already focused on. Local Whisper for offline privacy or hosted Whisper for 99+ languages — your call per session.
Where does my voice actually go?+
In local mode, nowhere. Audio is captured through cpal, transcribed in-process by whisper.cpp, and discarded. In cloud mode, audio is sent over HTTPS to the SpacialMind API and discarded after transcription. There is no third option.
Does it work in every app?+
Anywhere the OS accepts keyboard input. SpacialVoice uses Win32 SendInput on Windows, Quartz events on macOS, and wtype / xdotool on Linux — the same path the keyboard already takes, so editors, terminals, chat apps, and browsers all just work.
How do I get it?+
On-device transcription is included with the Basic plan. Cloud transcription (99+ languages) is included with Pro and Ultra. Download the installer from the buttons above, sign in, pick an engine, done.
Which Whisper models are available?+
Six ggml models locally: Tiny, Base, Small, Medium, and Large-v3 (English-only and multilingual variants where applicable). You download them on demand from inside the app; nothing is bundled. Cloud uses Whisper Large-v3 with multilingual auto-detect.
What hardware do I need?+
Any microphone. Built-in laptop mics work; a USB headset is better. CPU inference for Small fits comfortably on modern laptops; Medium/Large benefit from a GPU. The cloud engine doesn't care about your hardware.
Does it work offline?+
Local mode does — once a model is downloaded, you never need the network again. Cloud mode obviously needs it.
Which platforms are supported?+
Windows 10+, macOS 12+, and Linux (x11 and Wayland). Built on Tauri 2 with the same Rust core on every platform. macOS builds are notarized; Windows installers are NSIS .exe.
07 Access
Get SpacialVoice.
Local Whisper is included with Basic. Hosted multilingual Whisper unlocks at Pro.
Local dictation
Basic
$16/mo
$20/mo Save 20%
Get Started- SpacialMind account access
- SpacialDesk (ADE)
- Local Whisper on every Whisper model
- Custom dictionary + history
- 5,000 credits / month
- Email support
Adds the cloud engine
Pro
$40/mo
$50/mo Save 20%
Get Started- Everything in Basic
- Hosted Whisper for 99+ languages
- SpacialMemory (Live)
- SpacialMCP
- 12,500 credits / month
- Priority support
- Early access
Top-tier limits
Ultra
$80/mo
$100/mo Save 20%
Get Ultra- Everything in Pro
- 25,000 credits / month
- Highest usage ceilings
- Priority model routing
- Dedicated support
- Team seats (coming soon)
The rest of the stack
SpacialVoice slots into the SpacialMind toolkit. Use it on its own or alongside the rest of the family.
SpacialDesk
Vibe-coding workspace where terminals, tasks, code context, and AI agents share one focused desktop room.
SpacialMCP
Model Context Protocol server that gives your AI editor the ability to create projects, manage tasks, and configure agents.
SpacialCode
Agent-first desktop IDE with a multi-panel workspace and plan-based task execution — purpose-built for vibe coding.
Stop typing the easy stuff.
Download SpacialVoice and pipe your voice into anywhere a keyboard works.
7-day money-back guaranteeCancel anytime