# FluidVoice

> Free, open-source macOS dictation app with on-device AI enhancement via the Fluid-1 model — local-first, low-latency, and privacy-preserving.

FluidVoice is a free, open-source macOS dictation app built by Altic that pairs local speech-to-text transcription with Fluid-1, an on-device AI model for post-processing and smart formatting. The app is licensed under GPLv3 (from 2026-02-23 onward) and the project page reports over 50,000 downloads. It requires macOS 15.0 Sequoia or later and supports both Apple Silicon and Intel Macs.

## What It Is

FluidVoice sits in the voice-to-text dictation category, targeting Mac users who want fast, private speech input across any app without sending audio or text to the cloud. Unlike cloud-based dictation tools, FluidVoice processes everything locally using on-device speech models and an optional local AI enhancement layer. The core dictation experience works out of the box; Fluid Intelligence (the Fluid-1 model runtime) is a separate, optional ~3.5 GB download that adds smart formatting, context-aware capitalization, and adaptive tone without requiring API keys or network access.

## Two-Layer Architecture

The product is explicitly designed as two cooperating layers:

- **FluidVoice (the app):** Open-source macOS app handling the hotkey, overlay, speech model selection, system-wide text insertion via accessibility APIs, Command Mode, Write Mode, audio history, and settings.
- **Fluid-1 (the intelligence layer):** A privately maintained local AI runtime that cleans up rough dictation, adapts tone per app, formats structure, fixes casing, and handles dates, names, and numbers — all on-device. The README notes Fluid Intelligence is kept private to sustainably fund the free core experience, and this may change in the future.

## Speech Model Lineup

FluidVoice supports a broad range of on-device speech models, letting users trade off latency, language coverage, and download size:

- **Nemotron Speech 3.5** (Ultra Fast Low Latency and Multilingual) — ~40 languages, Apple Silicon only
- **Parakeet Flash (Beta)** — lowest-latency English dictation, ~250 MB, Apple Silicon
- **Parakeet TDT v3** — 25 languages, ~500 MB, Apple Silicon
- **Parakeet TDT v2** — English only, ~500 MB, Apple Silicon
- **Cohere Transcribe** — 14 languages, ~1.4 GB, Apple Silicon
- **Apple Speech** — zero-download, system languages, Apple Silicon + Intel
- **Whisper Tiny/Base/Small/Medium/Large** — up to 99 languages, ~75 MB–2.9 GB, Apple Silicon + Intel (Intel Macs supported via Whisper from v1.5.1+)

## Key Capabilities

- **Command Mode** — control macOS by voice: launch apps, run shortcuts, trigger system actions
- **Write Mode** — dictate or rewrite text inline in any text field across any app
- **Per-app tone profiles** — assign custom prompts per app so dictation adapts context (e.g., casual for Slack, formal for Mail, structured for GitHub issues)
- **AI post-processing** — optional cloud providers (OpenAI, Groq, custom) or local Fluid Intelligence
- **Live preview overlay** — notch-aware real-time transcription overlay
- **Audio history** — optional local recording history with ZIP export
- **Global hotkey** — instant voice capture from anywhere without switching apps
- **System-wide input** — text insertion into any app via accessibility APIs

The project page cites a real-time factor of up to 3,380x and perceived latency under 100ms, with support for multi-hour transcription sessions.

## Update: FluidVoice v1.6.1

The latest release is v1.6.1, published 2026-06-28. Version 1.6.0 introduced the rebuilt Parakeet implementation with near-zero delay, the Fluid Intelligence local AI model, adaptive light/dark theming, and a refreshed onboarding flow with language-first voice engine setup. The GitHub repository shows 4,946 stars and 303 forks as of early July 2026, with active development continuing toward planned iOS and Windows support.

## Features
- On-device speech-to-text transcription
- Fluid Intelligence local AI post-processing (Fluid-1 model)
- Command Mode — voice control of macOS
- Write Mode — dictate or rewrite text in any app
- Per-app tone profiles with custom prompts
- Global hotkey for instant voice capture
- System-wide text insertion via accessibility APIs
- Live preview overlay with notch support
- Multiple speech models: Nemotron, Parakeet, Cohere, Apple Speech, Whisper
- 40+ language support across models
- Optional cloud AI providers (OpenAI, Groq, custom)
- Audio history with ZIP export
- Today-usage stats
- Adaptive light/dark theming
- Menu bar integration
- Auto-updates with optional beta channel
- Local-first privacy — no data leaves Mac by default
- Optional anonymous analytics (opt-out available)
- Intel Mac support via Whisper models
- Apple Silicon optimized with CoreML and Metal

## Integrations
Cursor, Slack, Notion, Chrome, Discord, Ghostty, Terminal, Mail, Messages, VS Code, Xcode, Notes, Claude Code, OpenAI API, Groq API, Custom AI providers, Homebrew, GitHub Sponsors, FluidAudio SDK, NVIDIA Parakeet

## Platforms
WINDOWS, MACOS, LINUX, IOS, API, VSC_EXTENSION

## Pricing
Open Source

## Version
v1.6.1

## Links
- Website: https://altic.dev/fluid
- Documentation: https://github.com/altic-dev/FluidVoice
- Repository: https://github.com/altic-dev/FluidVoice
- EveryDev.ai: https://www.everydev.ai/tools/fluidvoice