wordbird 0.5.0

Voice dictation daemon using NVIDIA Parakeet on Apple Silicon

🦜 Wordbird

Contextual voice dictation for macOS. Powered by NVIDIA Parakeet running locally on Apple Silicon via MLX.

Press a hotkey, speak, and your words are transcribed and pasted into whatever app is focused. A small LLM post-processes the transcription to fix errors, using project-specific context from a WORDBIRD.md file.

Getting started

Requires macOS on Apple Silicon (M1+) and Python 3.10+.

# Run with uvx (no install needed)
uvx wordbird

# Or run in the background
uvx wordbird start
uvx wordbird stop
uvx wordbird status

Architecture

Wordbird runs as two sibling processes managed by a thin CLI:

• Server (wordbird-server) — FastAPI app handling transcription, post-processing, history, config, and serving the React dashboard
• Daemon (wordbird-daemon) — macOS-native process handling hotkeys, mic recording, overlay HUD, menu bar, and clipboard pasting

The daemon sends recorded audio to the server via HTTP. The server runs ML inference in a thread pool so the dashboard stays responsive during transcription.

uvx wordbird          # starts both (recommended)
uvx wordbird-server   # just the API server
uvx wordbird-daemon   # just the daemon (expects server running)

Context-aware correction

When dictating into Terminal.app, Wordbird detects the focused tab's working directory and looks for a WORDBIRD.md file up the directory tree. This lets you teach Wordbird your project's terms:

Context detection works with:

• Terminal.app — detects the focused tab's shell working directory
• VS Code / VS Code Insiders — via the Wordbird extension, which works with local and remote (SSH) workspaces

Transcription and pasting work in any app.

uvx wordbird init

This creates a WORDBIRD.md with the default prompt template. Edit it to add your project's terms:

---
transcription_model: mlx-community/parakeet-tdt-0.6b-v2
fix_model: mlx-community/Qwen2.5-1.5B-Instruct-4bit
---

{# Your correction prompt and examples here #}

{# Key terms: MyApp, some_function, PostgreSQL #}
{# Names: Alice, Bob #}
{# Misheard words: "bird word" should be "Birdword" #}

Input: "{{ transcript }}"
Output:

The file is a Jinja template. {{ transcript }} is replaced with the raw transcription. The YAML front matter lets you override models per-project.

Hotkey

Action	Default
Toggle recording	Right ⌘ + Space

Configurable via CLI flags or the dashboard settings:

--modifier-key KEY   Modifier key (default: rcmd). Options: rcmd, lcmd, ralt, lalt, rshift, lshift, rctrl, lctrl
--toggle-key KEY     Toggle key (default: space). Options: space, return, tab, escape

Options

--model MODEL        Transcription model (default: mlx-community/parakeet-tdt-0.6b-v2)
--fix-model MODEL    Post-processor model (default: mlx-community/Qwen2.5-1.5B-Instruct-4bit)
--no-fix             Disable LLM post-processing
--no-server          Don't spawn the API server (run it separately)

Dashboard

Wordbird runs a local web dashboard (default localhost:7870). Click the bird in the menu bar → Dashboard… to open it.

• History — browse transcriptions with timestamps, app name, working directory, and duration. See both original and corrected text.
• Settings — configure hotkey, models, and post-processing. Changes take effect within seconds.
• Stats — words dictated, recording time, WPM, session count.

uvx wordbird history        # view history from the CLI
uvx wordbird config         # show or create the config file

Data

All data is stored in ~/.wordbird/:

File	Purpose
wordbird.toml	User configuration
wordbird.db	Transcription history (SQLite)
server.json	Server port discovery
wordbird.pid	Singleton lock
wordbird.log	Background mode logs

Menu bar

Wordbird shows a bird icon in the menu bar:

• ⚪ White — idle
• 🟡 Yellow — connecting mic
• 🔴 Red — listening
• ✨ Sparkles — transcribing

Permissions

Wordbird needs three macOS permissions, granted to your terminal app:

• 🎤 Microphone — to record your voice
• 🔐 Accessibility — to paste text
• ⌨️ Input Monitoring — to detect the global hotkey

Wordbird checks these on startup and tells you what's missing.

Development

make backend-dev      # API server with hot reload
make daemon-dev       # daemon only (expects server running)
make frontend-dev     # Vite dev server with API proxy
make dev              # backend + frontend dev servers
make wordbird         # build frontend + run everything
make backend-test     # run pytest

License

MIT