# Juno

> A local, private voice layer for Mac that turns natural speech into clean text in any app, with no account, subscription, or cloud dependency required.

Juno is a native Mac application built by Cassini Research that converts spoken words into polished, finished text directly inside whatever app you are already using. It runs entirely on-device after a one-time model download, meaning audio, transcripts, and corrections never leave your Mac. The project is fully open source under the MIT license and requires no account, license key, or paid tier.

## What It Is

Juno is a local voice layer for Mac — a category that sits between basic OS dictation and cloud-based AI writing assistants. Where Apple Dictation inserts raw speech and cloud tools like Wispr Flow or Superwhisper send audio to remote servers, Juno processes everything on the Mac itself using Apple Silicon hardware. The core loop is: press a hotkey, speak naturally, watch a live HUD transcript, and receive cleaned final text inserted into the active app. It handles pauses, false starts, filler words, and mid-sentence corrections before committing the result.

## How the Voice Loop Works

Juno's architecture follows a single product path: audio → speech state → live preview → final transcript → writer → actions → native insertion. The Mac shell owns hotkeys, permissions, window state, secure-field policy, and text insertion. The local Python runtime owns speech processing, live preview, final transcription, writing cleanup, actions, dictionary and memory, app context, history, and health reporting. Key capabilities include:

- **Live HUD transcript** — words appear in a floating overlay while you are still speaking
- **Writing cleanup** — pauses, restarts, and take-backs are removed before text lands in the app
- **Smart formatting** — numbers, dates, prices, and URLs are normalized automatically (e.g., "nineteen ninety five" → 1995, "march twentieth at three p m" → March 20, 3:00 PM)
- **Selected-text rewrites** — spoken commands like "make this warmer" or "shorter" rewrite selected or recent text in place
- **Voice Actions** — "Hey Juno" triggers native Apple Notes, Reminders, and Calendar entries from a single utterance
- **Local vocabulary and snippets** — custom names, replacements, and saved phrases expand from voice

## Privacy Model

Juno is local-first by design. Microphone capture starts only when you trigger dictation; there is no background listening outside of an explicit "Hey Juno" mode. Runtime data — transcripts, corrections, dictionary, memory — is stored locally on the machine. Secure fields such as password boxes automatically suppress capture, context, history, recordings, learning, and paste. No account or sign-in is required, and history and recordings can be cleared at any time.

## Platform and System Requirements

Juno ships as a native macOS desktop application targeting Apple Silicon Macs. The homepage states minimum requirements of an Apple Silicon Mac, macOS 15 Sequoia or later, and 16 GB RAM. Running from source additionally requires Python 3.10 or newer, microphone permission, and Accessibility permission for native insertion. The GitHub README notes that Linux can run selected runtime checks and portable Python paths, but the shipping desktop product is macOS-first.

## Open-Source Architecture and Setup

The repository at github.com/Cassini-Research/Juno is MIT-licensed and written primarily in Python. Core folders include `juno_v2/` for the voice runtime, `juno_core_v3/` for broker contracts and model registry, and `shells/macos/` for the native Mac shell. Setup from source uses shell scripts: `bootstrap.sh` sets up the Python environment, `doctor.sh` checks requirements, and `run_live.sh` starts the local Mac voice stack. OTA updates are handled via a Sparkle-based release pipeline. The pre-built `.dmg` download is available from the website for users who do not want to build from source.

## Why It Got Attention

Juno launched on Product Hunt and positions itself as the only tool in its comparison set — alongside Apple Dictation, Wispr Flow, Superwhisper, and Aqua Voice — that combines a live HUD transcript, AI writing cleanup, Mac-native actions, selected-text rewrites, open-source code, and a permanently free model with no usage cap. The combination of full offline operation and zero subscription cost is the primary differentiator the project surfaces in its own comparison table.

## Features
- Local on-device speech processing (no cloud required after model download)
- Live HUD transcript while speaking
- AI writing cleanup removing pauses, filler words, and false starts
- Smart formatting for numbers, dates, prices, and URLs
- Selected-text rewrites via spoken commands (shorter, warmer, bullets, formal)
- Voice Actions for Apple Notes, Reminders, and Calendar
- Local vocabulary, snippets, and custom replacements
- Per-app writing context (email, chat, code, terminal)
- Privacy gates for secure/password fields
- No account, license key, or subscription required
- Fully open source (MIT license)
- Works offline after initial model download
- Clear history and recordings at any time

## Integrations
Apple Notes, Apple Reminders, Apple Calendar, Mail, Messages, Slack, Obsidian

## Platforms
MACOS, LINUX, API

## Pricing
Open Source

## Version
1.0.7-21

## Links
- Website: https://usejuno.co
- Documentation: https://docs.usejuno.co/docs
- Repository: https://github.com/Cassini-Research/Juno
- EveryDev.ai: https://www.everydev.ai/tools/juno-voice-layer