Fonos

# 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.

![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)
![Platform](https://img.shields.io/badge/platform-macOS%20%7C%20Linux-lightgrey)
![Built with Tauri](https://img.shields.io/badge/built%20with-Tauri%202-24C8DB)

![Fonos hotkey-first voice workflow](assets/hero/fonos-hotkey-hero.png)

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.

![Floating controls](assets/screenshots/floating-controls.png)

## Screenshots

| Dictation activity | Recent timeline |
|---|---|
| ![Dictation activity](assets/screenshots/dictation-result.png) | ![Recent timeline](assets/screenshots/recent.png) |

| Voice notes | Meeting summary |
|---|---|
| ![Voice notes](assets/screenshots/notes.png) | ![Meeting summary](assets/screenshots/meeting-summary.png) |

| Model setup |
|---|
| ![Model setup](assets/screenshots/models.png) |

## 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](https://github.com/ethannortharc/fonos/releases/latest), 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](https://github.com/ethannortharc/fonos/releases/latest):

```bash
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 wor

... (truncated)