Dictation that never leaves your Mac.

Hold fn, speak, release. Flowwy types what you said wherever your cursor is, in any app. Speech recognition runs on your Mac, an optional local model tidies the text, and coding agents get exactly the prompt you meant.

macOS 13+ · Apple silicon · free · also for Linux (beta)

The Flowwy Home screen: word count, words per minute and day streak, then today's dictations. Each entry shows the app it went to, the time, and the latency and model used.
The recording pill while listening: a coral lamp, a live waveform, a timer and buttons to cancel or finish.

What is Flowwy?

Flowwy is a free voice dictation app for Macs with Apple silicon. Hold a key, speak and release, and Flowwy types the text at your cursor in any app. Speech recognition runs on your Mac with whisper.cpp, so dictation works offline and your audio is never uploaded. An optional local AI model tidies the text, and coding agents such as Claude Code, Codex and Cursor get prompts in the format they expect. Flowwy is also available for Linux.

How Flowwy differs from cloud dictation
FlowwyCloud dictation
Where speech is recognizedOn your Mac (whisper.cpp on Metal)On the provider's servers
Audio uploadsNone. Audio stays in memory on your MacEvery dictation
Works offlineYes, after you download a speech modelNo

Flowwy also needs no account, costs nothing, and lists every network request it makes in Settings → Privacy.

How it works

Three steps, in any app with a text cursor.

  1. 01

    Hold your key

    Put the cursor anywhere: Mail, Slack, your editor, a terminal. Hold fn, or right ⌥, ⌘ or ⌃ if you prefer. Toggle mode keeps listening without holding.

  2. 02

    Speak

    The pill shows that Flowwy hears you. whisper.cpp transcribes on your Mac's GPU (Metal), with models from Whisper Tiny up to Large v3 Turbo.

  3. 03

    Release

    Let go and the text is inserted at the cursor. About 0.4 s from key release to text, measured on an M1 Max with a small model.

Text that fits where it lands

Flowwy listens to more than your voice: the app you're in, the words around the cursor, and the terms you taught it.

  • Context-aware

    Flowwy reads the text around the cursor and the app name, so names, casing and spacing fit what is already there. You can turn this off.

    app: Mail · before cursor: "Hi Jonas,"

  • AI cleanup, optional

    A local LLM fixes punctuation, removes filler words and self-corrections, and applies your style. If it takes too long, you get the raw transcript.

    • Ollama
    • llama.cpp
    • LM Studio
    • MLX
  • Dictionary

    Your names and jargon, plus misspelling → correction rules.

    kuber netties → Kubernetes

  • Snippets

    Say a trigger phrase and the full text is inserted.

    "my sign-off" → Thanks, talk soon. Sam

  • Styles

    Choose how your text reads, or keep the raw transcript.

    • Formal
    • Casual
    • very casual
    • Excited!
    • Raw
  • Scratchpad

    Voice notes for thoughts that don't belong in a text field yet.

  • Paste again

    Missed the field? A shortcut pastes your last transcript again.

Transforms

Select text. Press a shortcut. Done.

Transforms rewrite the selected text in any app and put the result back in place. The pill shows what changed.

  • Polish ⌥1 tightens and clarifies without changing what it says.
  • Prompt Engineer ⌥2 turns a rough request into a clear, structured prompt for an AI assistant.
  • Your own: write instructions, add examples, record a shortcut.

Transforms use your AI cleanup model: a local LLM, or Flowwy Cloud if you opt in.

The Transforms screen: a switch to use Transforms, the three steps select text, press ⌥1, review the changes in the pill, and cards for Polish (⌥1), Prompt Engineer (⌥2), Meeting summary, Stand-up and Create your own.

For developers

Talk to your coding agents

Flowwy knows when you're dictating into Claude Code in a terminal, a Codex composer or Cursor chat, and formats the text the way that tool expects.

Settings, Integrations: AI tools (MCP) for supporters, then Apps & agents with an explanation of slash commands, @file mentions and “and send”, and a list of CLI agents (Claude Code, Codex CLI, Gemini CLI, OpenCode, Aider, Cursor Agent, Amp and more), each with its profile and a status such as Installed or Not found.
You say
“slash compact”
Flowwy types
/compact
You say
“at source app dot tsx”
Flowwy types
@src/app.tsx
You say
“…fix the redirect loop and send”
Flowwy types
fix the redirect loop + Enter
You say
“rename user id”
Flowwy types
rename `userId`

Agents

Text is pasted as you said it, with no stray Enter. Ending with “and send” submits. File mentions resolve against the repository.

  • Claude Code
  • Codex
  • Gemini CLI
  • OpenCode
  • Aider
  • Cursor Agent
  • Amp
  • Goose
  • Crush
  • Claude
  • Codex app
  • T3 Code
  • Cursor
  • Supacode
  • Orca

Editors

Variable recognition matches spoken names to identifiers on screen. In Cursor and Windsurf chat, “tag index dot ts” inserts @index.ts, and matched names become inline code.

  • Cursor
  • VS Code
  • Windsurf
  • Zed
  • JetBrains IDEs
  • Xcode

Terminals and vocabulary

Long text is pasted in chunks so terminals keep up. Per-app profiles set style and cleanup. 13 vocabulary packs bias recognition toward the terms you use.

  • Terminal
  • iTerm2
  • Warp
  • Ghostty
  • kitty
  • WezTerm
  • Alacritty
  • Kubernetes
  • Terraform
  • AWS
  • GCP
  • Azure
  • Docker
  • Postgres
  • Kafka
  • Observability
  • Linux
  • TypeScript
  • Rust
  • Python

Bring your own model

The built-in whisper.cpp is the default. If you already run models locally, point Flowwy at them.

Default

Built-in whisper.cpp

Runs on
Your Mac's GPU (Metal)
Models
Whisper Tiny up to Large v3 Turbo
Setup
Download a model inside the app

Optional

Any local server

Runs on
An OpenAI-compatible server on localhost
Examples
Phonon-2 by Fermion Research, the whisper.cpp server
POST http://127.0.0.1:<port>/v1/audio/transcriptions
The Models screen, AI cleanup section: clean up text with a local model, run it built in (llama.cpp) or on a local server (Ollama, LM Studio, llama.cpp server or MLX), then the endpoint http://127.0.0.1:11434/v1, a timeout and the model, and a one-click Ollama preset.

Set up in a few minutes. Make it yours.

A guided setup walks you through permissions, your key and your first dictation. Then pick one of six themes, each in light and dark.

Guided setup, step 2 of 11: Allow three permissions. Microphone, Accessibility and Input Monitoring, each with a short reason, all three marked Allowed.

Interactive onboarding

Eleven short steps: pick a look, allow permissions, choose a speech model and your key, then try a real dictation, a dictionary word, a snippet and a style. The page updates as soon as macOS grants a permission.

  • Flowwy Home in the Soft theme, light appearance: serif numbers in cobalt on soft cards.
    Soft, light
  • Flowwy Home in the Soft theme, dark appearance: deep navy surfaces.
    Soft, dark
  • Flowwy Home in the Comic theme: ink-drawn panels, a pine sidebar and a hand-lettered headline.
    Comic

Also included: Clean, Pine, Paper and Midnight. Appearance follows macOS or stays light or dark.

Privacy you can check

Not a promise on a website. A number in the app's sidebar.

Audio, transcripts, your dictionary and the text around your cursor stay on your Mac. Recognition runs on-device and audio is kept in memory. The sidebar shows the network count all the time.

  • Settings → Privacy lists every request the app has made, with host, purpose and bytes.
  • Model downloads happen only when you click Download. The update check sends no audio, text or identifiers and can be turned off.
  • Speech and cleanup endpoints must be on this Mac. Remote addresses are refused unless you explicitly allow them.
  • No account, no telemetry. Flowwy Cloud is opt-in and shows up in the ledger when you use it.
Settings, Privacy: 0 bytes sent off this Mac, 602 MB downloaded for model files, the endpoints Flowwy can contact (speech server off, AI cleanup on loopback, the supporter licence only when you click, the update server off), and the network request log with time, host, purpose and bytes.

Flowwy Cloud · beta

No local model? Use ours, if you want.

Flowwy Cloud runs AI cleanup and Transforms on hosted models, so you don't have to run Ollama. It's optional, off by default and free during the beta.

  • Speech stays on your Mac

    Audio never leaves the Mac, in any plan. Only text is sent: the transcript (plus the app name and nearby text when context awareness is on), the selected text for Transforms, the note you summarize, and for Ask your question with the matching excerpts from your history and notes.

  • No text stored by Flowwy

    The gateway runs on Pixel & Process infrastructure in the EU and processes text in memory only; it logs token counts, never text. During the beta, requests are answered by third-party model providers that may be outside the EU, so use local cleanup for anything sensitive.

  • Opt-in per feature

    Turning it on allows only the Flowwy Cloud host. It doesn't open the app to other remote servers.

  • Falls back to local

    If the cloud can't answer, Flowwy uses your local model if one is set up. Every request appears in the ledger.

What's new

October 2026. Updates install from inside the app, signed and verified.

  1. New

    A softer look

    The new default theme, Soft, brings the iPhone app's look to the desktop: serif numbers, soft cards and navy buttons. Clean is still one click away.

  2. New

    Linux

    Flowwy now runs on Linux: .deb, .rpm and AppImage for x86_64 and ARM64, on X11 and Wayland.

  3. New

    Agent-aware dictation

    Slash commands, @file mentions and “…and send” for CLI agents and agent apps, per-app profiles and 13 developer vocabulary packs.

  4. Beta

    Flowwy Cloud

    Optional hosted AI cleanup and Transforms with device sign-in. Speech stays local.

  5. New

    Self-updates

    Flowwy checks for signed updates and installs them when you click. The check can be turned off.

  6. New

    Transforms and vibe coding

    Rewrite selected text with a shortcut. Variable recognition, file tagging and inline code in editors.

  7. New

    Guided setup

    An interactive onboarding in eleven steps, with a live key test and a first dictation.

  8. New

    Six themes

    Soft, Clean, Comic, Pine, Paper and Midnight, with light and dark appearance.

Questions

Does Flowwy send my voice or text anywhere?

No. Speech recognition runs on your Mac, and audio stays in memory. Local AI cleanup talks only to a server on your own Mac (loopback). Flowwy makes no network requests on its own except a check for app updates, which sends no audio or text and can be turned off. Every request is listed in Settings → Privacy.

Which Macs does Flowwy run on?

Macs with Apple silicon (M1 or later) running macOS 13 Ventura or later.

Can I dictate offline on a Mac?

Yes. Flowwy recognizes speech on your Mac with whisper.cpp on the GPU. Once you have downloaded a speech model inside the app, dictation, cleanup with a local model and Transforms work without an internet connection.

Is Flowwy a private alternative to Wispr Flow?

Flowwy does the same core job as other Mac dictation apps such as Wispr Flow: hold a key, speak, and clean text appears in any app. The difference is where the work happens. Flowwy recognizes speech on your Mac and never uploads audio. Cleanup runs on a local model unless you opt in to Flowwy Cloud, and the app lists every network request it makes.

Do I need an AI model for cleanup?

No. Dictation works with speech recognition alone. AI cleanup and Transforms are optional and use a local LLM server such as Ollama, llama.cpp, LM Studio or MLX, or Flowwy Cloud if you opt in.

What is Flowwy Cloud?

An optional, opt-in service in beta that runs AI cleanup and Transforms on hosted models, so you don't need to run a local LLM. Speech recognition always stays on your Mac. Flowwy's gateway does not store your text; during the beta, requests go to third-party model providers. It is free during the beta.

How do I dictate prompts into Claude Code, Codex or Cursor?

Put the cursor in the agent's input, hold your key and speak. Flowwy recognizes coding agents in terminals and agent apps: “slash compact” becomes /compact, “at source app dot tsx” becomes @src/app.tsx, and the text is pasted as you said it. Flowwy presses Enter only when you end with “and send”.

Is Flowwy available for iPhone, iPad, Windows or Linux?

Linux, yes: Flowwy comes as a .deb, an .rpm and an AppImage for x86_64 and ARM64. Windows, not yet. There is no iPhone or iPad version. To install Flowwy, open flowwy.dev on your Mac or Linux computer.

What does Flowwy cost?

The Flowwy app is free to download and use. Flowwy Cloud is free during its beta.

Which languages does Flowwy understand?

The multilingual Whisper models cover 99 languages and can detect the language automatically. Whisper Base (English) is a smaller English-only model.

Which permissions does Flowwy need?

Microphone to hear you while the key is down, Accessibility to type the text at your cursor, and Input Monitoring to notice your dictation key in any app. The guided setup asks for each one.

Download Flowwy

Free for macOS on Apple silicon. Once installed, Flowwy keeps itself up to date.

Flowwy runs on Macs with Apple silicon and on Linux. On a phone or tablet? Open flowwy.dev on your Mac or Linux computer to download it.

Install in five steps

  1. Download the DMG (20.5 MB).
  2. Open it and drag Flowwy to Applications.
  3. Start Flowwy. The guided setup asks for Microphone, Accessibility and Input Monitoring.
  4. Download a speech model and pick your dictation key.
  5. Hold the key, speak and release. The text appears at your cursor.
Version
latest
Download
20.5 MB
System
macOS 13 or later
Chip
Apple silicon
Speech model
78 MB – 1.6 GB
AI cleanup
optional
Price
free

Flowwy for Linux Beta

Free for x86_64 and ARM64. Pick the package for your distribution.

Arch Linux: use the AppImage for now. An AUR package (flowwy-bin) is coming. On ARM64, use the arm64 or aarch64 file name in the commands.

Version
latest
Desktop
X11 or Wayland
Chip
x86_64 (AVX2) or ARM64
.deb
Debian 12+, Ubuntu 22.04+
.rpm
Fedora 39+, openSUSE
AppImage
glibc 2.35+
Speech
on the CPU
Price
free
  • GNOME needs the AppIndicator extension for the tray icon (Ubuntu includes it). KDE, Xfce, Cinnamon and MATE show it as is.
  • On X11, hold Right Ctrl. On Wayland, the key goes through the desktop's Global Shortcuts portal (GNOME 48+, KDE Plasma 6; default Ctrl+Alt+Space), or, if you opt in, Flowwy reads the keyboard directly, which needs the input group. With the portal, Esc doesn't cancel, and on Wayland the Transforms and paste-again shortcuts work only while an X11 app has focus.
  • The AppImage updates itself. The .deb and .rpm update through your package manager: install the newer package the same way.