README

VoxCtrl Documentation

VoxCtrl is a high-performance, privacy-first voice-to-text dictation application and programmable voice input broker. All processing happens 100% on-device with zero telemetry or cloud dependencies.

---

Wiki Index

Document	Description
Overview	What VoxCtrl does, key features, and design principles
Architecture	System design, crate layout, data flow, concurrency model
Audio Pipeline	Audio capture, device management, VAD, resampling
Speech Recognition	Whisper engine, models, inference pipeline, post-processing
Routing	Output targets, hotkey bindings, delivery types
Hotkeys	Global hotkey listener, gestures, platform support
Text-to-Speech	TTS engines, voice packs, playback
Integrations	MCP server, DBus service, OpenAI-compatible LLM API, webhooks
UI & Windows	Svelte frontend, overlay, history viewer, settings
API Reference	Tauri IPC commands and frontend events
Configuration	All config files, schemas, and options
Installation & Setup	Dependencies, building, running
Development Guide	Dev environment, build system, crate structure

---

Quick Summary

Microphone → Audio Capture → Whisper Inference → Post-Processing → Output Router
                                                                         │
                                               ┌────────────────────────┤
                                               │                        │
                                          Inject text            Clipboard/File/
                                          to window              HTTP/Webhook/Socket/
                                                                 DBus/MCP/Exec/Pipe

Tech Stack:

• Frontend: Svelte 5 + Tailwind CSS 4 + Vite 5
• Desktop Shell: Tauri 2 (Rust + WebView)
• Backend: Rust (Tokio async), ~10 specialized crates
• Speech: whisper.cpp (GGUF models, CPU/CUDA/Vulkan)
• TTS: Piper (ONNX neural voices) + Espeak-ng fallback
• Config: TOML + JSON, hot-reloadable