Production-quality, local-first voice typing and AI text refinement for Windows.
Hold Ctrl+Space anywhere, speak naturally, and watch your polished words appear instantly.
- ποΈ Global Push-to-Talk (
Ctrl+Space) β Hold from any application (Notepad, VS Code, Word, Chrome, Terminal, Slack, Outlook), speak, release. - π Zero-Configuration Auto-Setup β Launches with a sleek animated splash banner. If no speech model is downloaded on launch, Vacanam automatically downloads and selects the Ultra Fast (
tiny~75 MB) model. - π£οΈ User-Centric Speech Engine Profiles:
β‘ Ultra Fast(tiny~75 MB) β Instant response for short phrases & low-end PCs.β Balanced (Recommended)(small~466 MB) β Optimal balance of speed and high accuracy.π― High Precision(medium~1.5 GB) β Exceptional accuracy for technical & complex terms.π Maximum Accuracy(large-v3~3.1 GB) β Maximum precision for multi-language speech & accents.
- β‘ Standalone Voice Action Commands & Shortcuts β Control your PC hands-free by speaking commands:
"select all"(Ctrl+A),"undo that"(Ctrl+Z),"redo that"(Ctrl+Y),"copy that","paste that","save file"(Ctrl+S),"press enter","press tab","press escape","delete line","delete word","switch window"(Alt+Tab),"close window"(Alt+F4), and"lock computer". - βοΈ Smart Verbal Punctuation & Emoji Formatting β Dictate natural spoken punctuation ("comma", "period", "question mark", "new line", "new paragraph", "open paren", "smiley face", "thumbs up", "fire emoji") and watch them seamlessly format with proper spacing and auto-capitalization.
- π Custom Voice Snippets & Expansion Macros β Define custom trigger phrases (e.g.
"insert signature","insert date","my meeting link") with dynamic{DATE},{TIME}, and{DATETIME}tags to expand boilerplate text instantly. - π€ Local LLM Text Refinement (AI Mode) β Sub-second CPU grammar polish using local GGUF models (
Qwen2.5-0.5B-Instruct&Llama-3.2-1B-Instruct) viaLLamaSharp. Custom System Prompts editable directly in Settings. - π Local SQLite Transcript History & Search β Opt-in local SQLite database (
%LOCALAPPDATA%\Vacanam\history.dbin WAL mode) saving timestamps, raw speech, polished text, and target app badges (devenv,notepad,chrome). Features instant real-time search, one-click Copy, and single-entry deletion. - π Microphone Health & Low Volume Alerts (<30%) β Real-time detection of muted microphones and low volume levels (<30%), warning you instantly on the floating overlay (
Mic Muted π/Mic Volume 20% π). - π¨ Windows 11 Fluent Dark UI β Animated Launch Banner, glassmorphic floating overlay with audio visualizer pulse ring, and dark Settings UI.
- π― Smart 3-Strategy Text Injection:
- Clipboard + Ctrl+V (Primary): Backup original clipboard β Paste β Restore original clipboard (<25ms).
- SendInput Unicode (Fallback 1): Native keystroke simulation for terminal windows (
cmd.exe,powershell,wt) without touching clipboard. - UI Automation (Fallback 2):
ValuePattern.SetValuefor accessible controls.
- π Zero-Trust Local Privacy:
- β No cloud APIs, no telemetry, no analytics
- β Audio is never saved to disk
- β Clipboard contents are never logged
- β Transcript history is opt-in (disabled by default) and stored 100% locally.
- π Read our full Privacy Policy or online at avikeid2007.github.io/Vacanam/privacy.html.
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β System Tray (Idle) β
ββββββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββ
β Hold Ctrl+Space
βΌ
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β WASAPI 16kHz PCM Capture + Active Window Context HWND β
ββββββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββ
β Release Ctrl+Space
βΌ
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β VAD Silence Trimming β Multi-Threaded Whisper.net β
ββββββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββ
β Transcribed Text
βΌ
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β (Optional) LLamaSharp GGUF Local LLM Text Polish β
ββββββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββ
β Final Text
βΌ
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Inject Text into Active App (Clipboard / SendInput) β
ββββββββββββββββββββββββββββββ¬βββββββββββββββββββββββββββββ
β (Opt-in)
βΌ
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Save Transcript Record to SQLite (%LOCALAPPDATA%) β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
- Download the latest installer
VacanamSetup-1.x.x.exefrom GitHub Releases. - Run
VacanamSetup-1.x.x.exe(no administrative privileges required). - Vacanam will launch automatically in your System Tray β hold
Ctrl+Spaceanywhere to dictate.
Note
Code Signing: Free code signing provided by SignPath.io, certificate by SignPath Foundation.
- Windows 10 / 11 (x64)
- .NET 10 SDK β Download .NET 10
# Clone the repository
git clone https://github.com/your-org/vacanam.git
cd Vacanam
# Restore dependencies
dotnet restore Vacanam.slnx
# Build Release binary
dotnet build Vacanam.slnx --configuration Release
# Run Vacanam (runs in System Tray)
dotnet run --project src/Vacanam.App --configuration ReleaseRight-click the Vacanam tray icon and click Settings:
- Speech Tab:
- User-Centric Profiles: Select from
Ultra Fast,Balanced,High Precision, orMaximum Accuracy. - Live Model Badges: See which models are
ACTIVE & READY,DOWNLOADED, orNOT DOWNLOADED. - One-Click Download & Select: Download models directly from Settings with live progress tracking.
- User-Centric Profiles: Select from
- Audio Tab: Live Microphone Volume Slider (0 - 100%), Mute Microphone Toggle, and Voice Activity Detection (VAD) threshold settings.
- AI Tab: Enable/disable AI text polish, customize System Prompt / Grammar Rules, and manage local GGUF models (
Qwen 2.5 0.5B&Llama 3.2 1B). - History Tab: Local SQLite Transcript Search, instant real-time filtering, target application badges, one-click Copy, single-record deletion, and Clear All History.
- General Tab: Startup & notification options.
- Privacy Tab: Local-first guarantees & opt-in transcript history toggles.
Vacanam is built with a strictly decoupled modular architecture across 8 projects:
Vacanam.slnx
βββ src/
β βββ Vacanam.Core/ Pure domain interfaces, models, and enums (net10.0)
β βββ Vacanam.Windows/ Win32 P/Invoke (RegisterHotKey, GetForegroundWindow, SendInput)
β βββ Vacanam.Audio/ WASAPI audio capture, pure C# AudioConverter, VAD
β βββ Vacanam.Speech/ Whisper.net speech recognition & automatic model downloader
β βββ Vacanam.LLM/ LLamaSharp v0.27.0 CPU local LLM text refinement engine
β βββ Vacanam.Input/ Text injection strategies (Clipboard backup/restore, SendInput, UIA)
β βββ Vacanam.Infrastructure/ Settings persistence (JSON), SQLite history, DI registration
β βββ Vacanam.App/ WPF UI (Tray, Floating Overlay, Launch Banner, Settings Window)
Models, settings, databases, and operational logs are stored in your local application data directory:
%LOCALAPPDATA%\Vacanam\
βββ Models\
β βββ Whisper\
β β βββ ggml-tiny.bin (Ultra Fast ~75 MB)
β β βββ ggml-small.bin (Balanced ~466 MB, Recommended)
β βββ LLM\
β βββ Qwen2.5-0.5B-Instruct-Q4_K_M.gguf (~398 MB)
β βββ Llama-3.2-1B-Instruct-Q4_K_M.gguf (~808 MB)
βββ history.db (SQLite database - opt-in history)
βββ Logs\
β βββ vacanam-YYYYMMDD.log
βββ settings.json
Distributed under the MIT License. Free and open-source forever.
Vacanam is an independent project designed for local offline voice typing on Windows.