ideaplexa/voicetypr

Voicetypr - AI powered offline voice to text dictation tool for busy founders, vibe coders, AI power users on macos, windows. Alternative to wispr flow and superwhisper.

What is Voicetypr?

Voicetypr is an offline‑first desktop dictation tool for macOS and Windows. It lets you speak anywhere on your computer, transcribes the audio with local or optional cloud speech‑to‑text models, and inserts the resulting text right at the cursor.

Core ideas (from the README)

Idea	What it means
System‑wide dictation	A global hot‑key starts/stops recording, so you can dictate into any app (email, code editor, browser, etc.).
Local transcription by default	Uses OpenAI‑Whisper (macOS & Windows) and Apple‑Silicon‑optimized Parakeet models. No internet traffic unless you choose a cloud provider.
Optional cloud providers	If you prefer a different service you can point Voicetypr at Soniox, OpenAI, Groq, Deepgram or Cohere.
AI formatting	After transcription the raw text can be cleaned‑up by a LLM (OpenAI, Anthropic, Gemini or any OpenAI‑compatible endpoint).
CLI & agent integration	A `voicetypr` command line tool gives scripts or autonomous agents access to the same transcription pipeline, returning plain text or structured JSON.
Network sharing	One Voicetypr instance can act as a private transcription server on your LAN, letting other installations off‑load work.
Lifetime license for local models	After a trial you can buy a one‑time license that removes any per‑minute fees for the locally‑run models.

Main features (as listed)

Global shortcuts, push‑to‑talk, toggle recording, auto‑insertion at cursor
Whisper (macOS/Windows) + Parakeet (Apple Silicon) for on‑device transcription
Cloud STT options (Soniox, OpenAI, Groq, Deepgram, Cohere)
LLM‑based text polishing via OpenAI/Anthropic/Gemini or custom endpoint
Transcribe audio/video files; cloud providers can add speaker diarization
History view with search, metadata, compare raw vs. formatted, copy or re‑transcribe
LAN‑based private transcription server (network sharing)
Agent‑ready CLI (voicetypr transcribe … --json etc.)
Stable/Beta update channels; Microsoft Store version respects Store updates
Small React UI backed by Rust for audio capture, hotkeys, and cursor insertion

How you use it

Install – macOS 14+ DMG or Windows installer / Microsoft Store.
Grant permissions – microphone and (macOS) accessibility for cursor insertion.
Download a model – the app fetches Whisper/Parakeet files once and stores them locally.
Set a hot‑key – choose a global shortcut in Settings → General.
Dictate – place the cursor in any text field, hit the shortcut, speak, stop.
Result – Voicetypr types the transcript at the cursor and saves it to the history.

Command‑line interface (CLI)

The CLI mirrors the GUI capabilities for scripts or AI agents:

voicetypr status --json          # health info
voicetypr models --json          # list available local/cloud models
voicetypr transcribe --file note.wav --json   # file → text/JSON
voicetypr record --until-silence --json       # live mic capture

Output is human‑readable by default; --json yields a structured object suitable for automation. Audio never leaves the machine unless you explicitly request a remote server or cloud provider.

Privacy & data flow

Mode	What leaves the computer
Local transcription	Nothing – audio and text stay on‑device.
Cloud transcription	Recorded audio is sent to the chosen STT provider.
AI formatting	The raw transcript is sent to the configured LLM service.
Network sharing	Audio is sent to the Voicetypr server you configured (LAN or other network).
Diagnostics and analytics can be toggled in Settings; the project’s privacy policy details retention.

Technical stack

Desktop shell: Tauri v2 (Rust + webview) for native windowing, auto‑updates, permissions.
Frontend: React 19, TypeScript, Tailwind CSS, shadcn/ui, Zustand state management.
Backend: Rust handling audio capture, resampling, hot‑keys, transcription orchestration, history, cursor insertion.
Local engines: Whisper (CPU) on both OSes; Parakeet side‑car on Apple Silicon; optional Vulkan‑accelerated Whisper side‑car on Windows.
Network sharing: Authenticated client/server built on the same Rust backend.

Getting the code & building yourself

git clone https://github.com/ideaplexa/voicetypr.git
cd voicetypr
pnpm install               # Node + React deps
pnpm tauri:dev             # Run in dev mode (requires Rust toolchain, Tauri prerequisites)

Prereqs: Node + pnpm, Rust stable, Tauri v2 platform tools, Xcode CLI (macOS) or Visual Studio Build Tools (Windows). Lint, type‑check, and test scripts are provided (pnpm lint, pnpm test, etc.).

License & community

License: GNU Affero General Public License v3 (AGPL‑3.0).
Support: GitHub Issues, in‑app “Report a problem”, and a public website with docs, changelog, and help pages.
Contributing: Follow the guidelines in AGENTS.md and CLAUDE.md; keep platform‑specific behavior explicit and maintain the CPU‑safe main process on Windows.

Bottom line

Voicetypr is a fully open‑source, cross‑platform dictation app that prioritises privacy by running Whisper‑based speech‑to‑text locally, while still offering cloud and LLM‑based formatting as optional add‑ons. Its CLI makes it useful for automation and AI‑agent workflows, and the architecture (Rust core + React UI) gives native performance with a modern UI.