GitHub - AbhishekBarali/SpeakoFlow: Free, open-source offline voice dictation for Windows, macOS, and Linux. A Wispr Flow alternative with an AI assistant that can read your screen on request and answer questions.

English · 简体中文 · 繁體中文

Talk and it types, in any app. Ask and it answers. Start a conversation and talk it through.
It also writes up your meetings. Free, open source, and local by default.

Website  ·  Documentation  ·  All releases

What it does

You think faster than you type. SpeakoFlow lets you work with your voice instead, from any app, with a keyboard shortcut:

_{Windows defaults. The macOS and Linux keys are under Keyboard shortcuts, and you can change every one.}

Speech is transcribed on your own computer unless you pick a cloud service. The assistant uses whichever model you choose: a built-in one that works offline, your own Ollama or LM Studio server, or a cloud provider with your own API key.

I built SpeakoFlow while studying alone for exams. I was paying for a dictation app that could hear me but couldn't help me, so I made one that does both.

Features

Dictation

Hold the keys, talk, and let go. Your words appear wherever the cursor is, and the recording pill can show them as you speak. Transcription runs on your graphics card or processor with a local model: Parakeet by default for English, Nemotron for 28 languages with automatic detection, Whisper for 99, and 69 speech models in the catalog altogether, which you can filter by language. If you'd rather use the cloud, ElevenLabs, Deepgram, and Amazon Transcribe stream text while you talk, and OpenAI, Groq, Mistral, Azure AI Speech, and OpenRouter work too.

• Undo. Cancelled a recording by accident? The pill offers Undo for a few seconds, and History keeps the recording so you can recover it later. A failed transcription offers Try again.
• Your words, spelled your way. Add names and jargon to the Dictionary, or write text replacements. On Windows it can learn the words you correct.
• Translate to English. Whisper, Canary, Granite Speech, and Voxtral models, and OpenAI or Groq in the cloud, can turn speech in another language into English text.
• Hold, tap, or go hands-free. Hold the keys while you talk, or switch to tap so one press starts and the next one stops, separately for dictation and the assistant. Holding to dictate and need longer? Press Space too, let go, and keep talking.

Ask

Select some text, or don't. Hold Left Ctrl + Left Alt and say what you want: "translate this to Spanish", "write a polite reply saying I can't make Thursday", "explain this". The answer streams into a card over the app you're in, and you can copy it, insert it at the cursor, or put it in place of the text you selected.

It can do more if you let it:

• Screen vision. Ask about the error in your terminal or the chart in your spreadsheet. It's off until you switch it on, and even then the model decides per question whether it needs to look. A screenshot it doesn't use never leaves your computer.
• Web search through Serper, Brave, Tavily, Exa, SerpAPI, or TinyFish.
• Reminders. "Remind me to send the invoice in twenty minutes." Reminders survive a restart and pop up without taking your keyboard.
• Profiles and memory. Give it different personas, each with its own reply length, and let it remember how you like to work. Memory is off by default, stays on your computer, and you can edit or erase it.

Conversation

Press Left Ctrl + Left Alt + C and talk. The assistant answers out loud, and you can cut in while it's speaking. Esc stops a reply without ending the conversation, and pressing the keys again ends it. Need to type something in the middle? Dictate as usual. The conversation waits while you do and picks up again afterwards.

Replies are spoken by a voice on your computer (Kokoro, Kitten, Pocket TTS, or Supertonic) or by a cloud voice from OpenAI, ElevenLabs, Deepgram, Cartesia, Google, Azure, and others.

Meetings ^beta

Press Start recording before a call. SpeakoFlow records your microphone and your computer's audio as two streams and transcribes them as people speak, so it always knows which words were yours. No bot joins the meeting, and it works with any meeting app. On Windows it can offer to record when it notices a call.

When the call ends it writes the notes, with a summary, decisions, and next steps with owners, using a template you pick (General, Standup, One-on-one, Interview, or Action items). Other voices are labelled Speaker 1, Speaker 2, and so on. Afterwards you can ask questions about the meeting, or choose Discuss this meeting to talk it over in a conversation.

On macOS, recording the other side of a call needs a virtual audio device such as BlackHole. See Troubleshooting.

AI cleanup

SpeakoFlow Mini is a small model we trained for one job: turning what you said into clean text. It removes filler words, fixes grammar and punctuation, and follows spoken edits, so "scratch that" or "actually, eleven" does what you meant instead of being typed out. It's a 795 MB download, runs on your computer, and handles English for now. Any other local or cloud model can do the job instead, including Apple Intelligence on Apple silicon Macs.

Cleanup is off until you turn it on. It then gets its own shortcut (the dictation keys plus Shift) or runs on every dictation. On top of it you can add a writing style: Professional, Friendly, Concise, Formal, Casual, or one you write yourself.

Also in the app

• Insights. Words dictated, your speaking speed, time saved compared with typing at 40 words a minute, questions asked and time on calls with the assistant, meetings recorded, and six months of activity with streaks. Share turns it into a set of cards, one insight each, to copy, save, post to X or LinkedIn, or send through your system's share sheet, in square, story or wide; they are made on your computer and nothing is uploaded.
• History. Your dictations, questions, and conversations. Play a recording back, transcribe it again, recover a dismissed one, or continue a chat as a conversation. Old recordings can delete themselves after a set number, days, or months.
• Models you already have. Add a .gguf or Whisper .bin file, or link a folder and every model in it shows up. Nothing is copied or moved. Downloads that do happen fetch eight chunks at once and resume where they stopped.
• Generate with Flow. Start a dictation with "Hey Flow" and describe what you want written, and the finished text is pasted instead of your words. It's off by default, under Settings → Dictation.
• 20 interface languages.

Each feature has its own page in the documentation.

A look around

Keyboard shortcuts

¹ Press Space while you hold the dictation keys, or all together. It keeps listening after you let go, until you press the dictation keys again. The first few times you hold the dictation keys, the pill shows the Space key as a reminder. Pick another key, or turn it off, under Hands-free in Settings → Shortcuts.

² Space is already part of Ctrl + Space, so turn off Hold to talk for dictation instead, which makes every dictation hands-free.

³ Only while AI cleanup is on and has its own shortcut.

The pattern is the same on every platform. Add Shift to the dictation keys to dictate and clean up, and add C to the ask keys to start a conversation. Recording shortcuts work while you hold them. Dictation and the assistant each have their own Hold to talk switch in Settings → Shortcuts (Manage on the Home page): turned off, one press starts and the next one stops. A dictation that keeps listening without the keys is hands-free: the pill gets a ring, and hovering it shows ✕ to cancel and ✓ to finish.

Esc only cancels while something is running, like a recording or a reply being read aloud, so other apps keep their Esc the rest of the time. To change a shortcut, click its keys. Cancel, the conversation shortcut and the hands-free key can also be turned off from there.

On a Mac, set System Settings → Keyboard → Press 🌐 key to to Do Nothing, or the globe key opens the emoji picker as well. Macs that were on the older Option + Space default keep it after updating.

Scripts and window managers can control SpeakoFlow with command-line flags such as --toggle-transcription.

Models and providers

Every job can run on your computer or with a provider you choose. Cloud providers use your own API key, stored in your system keychain.

Privacy

By default your voice is transcribed on your computer and never uploaded. There is no telemetry, no analytics, and no account.

Data leaves your computer only for things you set up yourself:

• A cloud speech service, if you pick one instead of a local model. It receives your recordings.
• The assistant's provider, if it isn't a local one. It receives your questions, any text you selected, a screenshot when screen vision is on and the model asks for one, and a meeting's transcript when you generate notes or ask about it.
• Web search, if you turn it on. The search provider receives the query.
• Feedback, if you send it from the app. It sends exactly what the dialog shows you, to a private issue tracker only the developer can read.

API keys live in your system keychain. Memory is off until you turn it on, and it stays on your computer where you can view, edit, or erase it. More detail is on the privacy page.

Install

Download the latest build from the Releases page. On first launch you pick a speech model, and a short tour shows the shortcuts while it downloads.

Already on 1.4 or earlier? Those versions can't update themselves to 2.0, so download 2.0 once from Releases and install it over the old one. Your settings and history are kept. From 2.0 on, updates install from inside the app.

Windows

Run the .exe installer. Windows may show a SmartScreen notice because the installer isn't signed by a known publisher yet; choose More info → Run anyway.

macOS

Download the .dmg for your Mac (aarch64 for Apple silicon, x64 for Intel) and drag SpeakoFlow into Applications. The app isn't signed by Apple yet, so macOS says it "is damaged and can't be opened". It isn't damaged. Clear the block once with this command in Terminal, then open the app normally:

xattr -dr com.apple.quarantine /Applications/SpeakoFlow.app

SpeakoFlow then asks for Microphone and Accessibility permission so it can hear you and type into other apps.

Linux

• Arch Linux. Install speakoflow-bin from the AUR, for example with yay -S speakoflow-bin.
• Debian 13+, Ubuntu 24.04+, Mint 22+, Pop!_OS. Install the .deb, which also adds the app icon and menu entry:

  sudo apt install ./SpeakoFlow_*_amd64.deb

• Other distributions, including Fedora and openSUSE. Use the AppImage: make it executable with chmod +x and run it. Tools like Gear Lever or AppImageLauncher add it to your app menu.

Both packages are built for x86_64 and ARM64 on Ubuntu 24.04, so they need glibc 2.39 or newer. That rules out Ubuntu 22.04, Debian 12, Mint 21, and RHEL 9 and its rebuilds. There's no .rpm yet, because the packaging doesn't bundle the speech engine correctly, and a package that installs but can't transcribe would be worse than none.

Updates

SpeakoFlow checks for new versions in the background and installs them from Settings → About, after verifying each one against the project's signing key. The AUR package updates through your package manager instead. To hear about releases on GitHub, click Watch → Custom → Releases at the top of this page.

Build from source

You need Rust and Bun.

git clone https://github.com/AbhishekBarali/SpeakoFlow.git
cd SpeakoFlow
bun install
mkdir -p src-tauri/resources/models
curl -o src-tauri/resources/models/silero_vad_v4.onnx https://blob.handy.computer/silero_vad_v4.onnx
bun run tauri dev

On Arch-based distributions, bun run install:arch builds the current checkout and installs it under ~/.local with a desktop entry and a speak command. BUILD.md has the setup for each platform.

The app is Tauri 2 with a Rust backend and a React and TypeScript frontend. Speech runs on transcribe.cpp, which loads every local model, Whisper .bin files included, and voice detection on Silero VAD with ONNX Runtime; the assistant and cleanup on a bundled llama.cpp engine or any OpenAI-compatible API; local voices on Kokoro in the app's window and sherpa-onnx on the processor; and meeting speaker labels on WeSpeaker voice embeddings.

How it compares

_{Prices and platforms from each product's own site, checked October 2026.}

SpeakoFlow's dictation core comes from Handy, which is a good choice if dictation is all you need. More detail: SpeakoFlow vs Wispr Flow and free and open-source Wispr Flow alternatives.

Troubleshooting

The common problems are below. For anything else, see the troubleshooting docs or open an issue.

Roadmap

Where SpeakoFlow is heading, written down here so the project's direction is public and dated. These are ideas, not promises: the order isn't a schedule, some will change shape, and some will be dropped. If one matters to you, say so in an issue.

Last revised October 2026.

Soon

• Code signing for Windows and macOS, so installs stop warning.
• More one-click local models, and smaller, faster ones for laptops without a graphics card.
• More community translations of the interface.
• Smart spacing and AI cleanup for more languages. SpeakoFlow Mini handles English today.

Voice commands and agents

• Voice commands that take actions. Say "open the terminal", "switch to Chrome", "mute", "take a screenshot", "turn the volume down" and it happens, instead of ending up in a text field.
• Editing by voice. "New line", "delete that", "select the last sentence", "make that bold", "send it", spoken while you dictate, act on the field you're in instead of being typed out. Cleanup already understands "scratch that"; this extends it to the whole field.
• An agent that finishes a task for you. Describe a job out loud, such as "put these receipts into a spreadsheet" or "reply to everyone from today's meeting", and the assistant works through it on your computer with screen vision, the keyboard, and the mouse. It shows each step, asks before anything it can't undo, and stops the moment you tell it to.
• Tools the agent can use. Browser, files, calendar, email, terminal, and the apps you choose. Every tool is off until you switch it on, each with its own permissions, and you can see what each one did.
• Routines. Several steps under one phrase: "start my standup" opens your notes, starts the meeting recording, and picks the template.
• Dictation tuned for agentic coding. Talking to coding agents and terminals: file paths, symbols, and commands spelled right, and a prompt mode that turns a rambling description into a tidy instruction.
• Help writing prompts. Describe what you want to build and get a solid prompt back, for any model or tool.
• Works with other agents. Use MCP servers as tools, and let other agents on your computer dictate, listen, and speak through SpeakoFlow over a local API.

Ask and Conversation

• A memory that lasts. Longer memory, memory per profile, and a plain view of everything it remembers.
• Ask about your files. Point it at a folder and ask questions about what's in it, with the index kept on your computer.
• Watch mode. A conversation that keeps an eye on the screen while you work and speaks up when you ask it to.
• Wake word for the assistant. "Hey Flow" already starts Generate with Flow; let it start a question or a conversation too, without the keys.
• More voices and more languages for spoken replies, local first.

Meetings

• Out of beta, with calendar awareness, speaker names it can learn, and search across all your meetings.
• Live captions and translation during a call.
• Send notes where they go: Notion, Obsidian, Google Docs, email, or a folder of Markdown.

A cloud option

Local stays the default, and the app will always work with no account at all. The cloud is for people who want it.

• Sync. An optional account that keeps your Dictionary, profiles, settings, and history the same on every computer, encrypted so only you can read it.
• SpeakoFlow Cloud as a provider. Transcription, the assistant, and voices from one place with one subscription, for people who'd rather not manage API keys or run models locally. Pick it like any other provider.
• Teams. Shared dictionaries, styles, and profiles for a company, with an admin who sets which providers are allowed.
• A phone companion. Dictate on your phone and have it typed on your computer, or use the assistant on the go.

Platform

• More ways to install: Microsoft Store, Homebrew, Flatpak, and an .rpm.
• Better Linux, especially on Wayland.
• Faster on more hardware: NPUs on Windows laptops and the Neural Engine on Macs, with lower latency throughout.
• Push-to-talk from other devices: foot pedals, Bluetooth buttons, headset and earbud controls, Stream Deck.
• Fully usable by voice and screen reader, for people who can't use a keyboard or a mouse.
• An extension API, so others can add tools, providers, cleanup styles, and commands.

Contributing

Contributions are welcome. CONTRIBUTING.md explains how to get started, and CONTRIBUTING_TRANSLATIONS.md covers translating the app.

Found a bug or have an idea? Use Send feedback in the app (the ? next to Settings), or open an issue.

License and credits

SpeakoFlow is released under the MIT License.

The dictation core comes from Handy by CJ Pais, used under the MIT licence. Thanks to CJ for making it open. The assistant, conversations, meetings, screen vision, Generate with Flow, translation, spoken replies, and memory are SpeakoFlow's own.

Thanks also to Tauri, transcribe.cpp, llama.cpp, ONNX Runtime, sherpa-onnx, Silero VAD, WeSpeaker, Kokoro, and Kyutai's Pocket TTS.

Cloud transcription uploads are compressed with the LAME MP3 encoder, via mp3lame-encoder. Both are LGPL-3.0 and are statically linked; their source, and this app's, are public, so a build against a modified LAME is always possible.