Fonos

A voice terminal for macOS and Linux. Inspired by Whisper Flow and Taplines, Fonos lets you hold a hotkey, speak naturally, and send the result to the place you actually need it: the cursor, clipboard, a notebook, a meeting transcript, text-to-speech, or an AI agent.

Fonos is open source, local-first, and provider-agnostic. Bring your own API keys for OpenAI, Anthropic, Google, OpenRouter, and more, or run locally with Ollama / LM Studio. Your keys, transcripts, notebooks, and meeting notes stay on your machine.

What it does

• Dictate anywhere: press-and-hold to record, release to transcribe, then optionally clean, rewrite, translate, or paste.
• Stay out of the main window: global hotkeys summon floating controls, so most captures happen without opening the full app.
• Turn speech into structured notes: route voice captures into notebooks with custom prompts and export to Markdown or JSON.
• Capture meetings: record mic + system audio as separate speaker channels, then generate an AI summary with decisions and action items.
• Configure your own model stack: mix cloud and local STT, LLM, and TTS providers per mode, notebook, or workflow.
• Use your voice as an agent input: talk to an AI agent with allow/block lists for tool execution.

Hotkey-first floating controls

Fonos is designed to live under your shortcuts, not as a window you constantly manage. Use the global hotkeys to bring up small, contextual surfaces:

• Float pill: appears near the cursor for dictation, recording state, processing state, and quick mode switching.
• Mode roller: switch between Raw, Polish, Translate, Note, or custom modes without opening Settings.
• Note panel: pick a notebook and capture directly into it from a compact floating panel.
• Agent panel: talk to the agent or act on selected text, with tool execution shown inline.

Screenshots

Dictation activity	Recent timeline

Voice notes	Meeting summary

Model setup

Features

🎙️ Dictation

Press-and-hold to record, release to process. Each mode is a transcription plus an optional LLM pass with its own prompt.

• Built-in modes: Raw (verbatim), Polish (natural writing), Formal, Translate, Clean (filler removal).
• Unlimited custom modes — your own system prompt, template, temperature, model override, and auto-paste behavior.
• Live activity feed with model + latency, animated waveform, and a horizontal mode selector.

📓 Voice Notes

Organize recordings into notebooks.

• Quick Note catch-all plus unlimited custom notebooks, each with its own STT/LLM model, processing mode, and prompt.
• Bind up to 3 notebooks to dedicated hotkeys; capture from a compact floating panel.
• Export any notebook to Markdown or JSON.

👥 Meeting Capture

Real-time transcription with separate speaker channels.

• Dual capture — your mic and system audio (remote participants, via ScreenCaptureKit) are transcribed independently.
• Live timestamped transcript panel, plus an AI summary with key points, decisions, and checkable action items when you stop.

🤖 Agent

Voice-driven AI conversations with tool execution.

• Hold a hotkey to talk; responses stream with thinking indicators.
• Skill execution with allow/block command lists; optional spoken (TTS) replies.

🔊 Voice (TTS & cloning)

• Type or paste text, pick a voice, adjust speed (0.5×–2×), and synthesize.
• Clone a voice from a 3–10s recording or an uploaded clip.

📊 Stats & Recent

• Usage over 7 / 30 / 90 days: words dictated, session counts, estimated time saved.
• A unified, filterable timeline across dictation, agent, notes, and meetings.

Install

macOS

Download the latest .dmg from Releases, open it, and drag Fonos to Applications. Apple Silicon, macOS 13.0+. Signed and notarized — no Gatekeeper workaround needed.

Linux

Download the .deb or .rpm (amd64 / arm64) from Releases:

sudo apt install ./fonos_*.deb    # Debian / Ubuntu
sudo dnf install ./fonos-*.rpm    # Fedora / RHEL

Text injection (paste-at-cursor) needs xdotool (sudo apt install xdotool). On Wayland it works via XWayland.

Build from source

Prerequisites

• Rust (stable) and the Tauri CLI — cargo install tauri-cli --version "^2"
• Node.js 20+
• macOS: Xcode Command Line Tools (xcode-select --install) — provides swiftc for the Speech / ScreenCaptureKit helpers
• Linux: the system packages listed in .github/workflows/build-linux.yml (libwebkit2gtk-4.1-dev, libgtk-3-dev, libasound2-dev, …)

Run it

git clone https://github.com/ethannortharc/fonos.git
cd fonos/fonos-desktop
npm ci
cargo tauri dev          # hot-reloading dev build

Package a release — .app / .dmg on macOS, .deb / .rpm on Linux:

cargo tauri build

The compiled macOS helper binaries are checked in, so builds work out of the box. To rebuild them after editing the Swift sources:

./src-tauri/swift/build.sh   # macOS only

Providers

Configure any mix in Settings → Models — set global defaults, then override per-mode or per-notebook.

STT supports two API paths: Whisper multipart upload, or chat-completions base64 audio for multimodal models.

Keyboard shortcuts

All remappable in Settings → Hotkeys.

Repository layout

Fonos is a monorepo. This README focuses on the desktop app.

Tech stack

Desktop: Tauri 2 · Rust · React 19 · TypeScript · Vite · Tailwind CSS · SQLite (rusqlite). macOS speech and system-audio capture run through small Swift helpers built on Speech and ScreenCaptureKit.

Contributing

Issues and pull requests are welcome. Run the tests with:

cargo test                              # Rust (core + desktop)
cd fonos-desktop && npm run test:e2e    # Playwright end-to-end

Some desktop tests need microphone / accessibility permissions. Please keep changes focused and match the surrounding style.

License

MIT © Ethan