Type with your voice.
Hold a key and talk. Your words appear in whatever app you’re using.
That's the Apple Silicon build. On an Intel Mac, take the Intel disk image. Either way, one Terminal command clears the "damaged" warning. How to
Fourteen apps stand on a turntable around one text field: Slack, Gmail, ChatGPT, Notion, Linear, WhatsApp, Cursor, Google Docs, Claude, Discord, GitHub, Obsidian, Zoom, and Telegram. It turns one app at a time to the front, where SpeakoFlow's recording pill listens and the sentence is typed into the field and dropped into the app, such as "Running late, start without me" in Slack.
Dictate in any language.
Up to 99 languages, typed wherever your cursor is.
Hold CtrlWin and speak
The same note is dictated in English, Spanish, French, German, Hindi and Japanese, each with a different speech model: Parakeet, Whisper and Canary on the computer, Deepgram, ElevenLabs and Groq in the cloud. In live transcription the words show in the recording card as they are said; in normal transcription a small pill listens and the finished text appears when you stop.
Ask for anything.
Select some text, or don't, and say what you need. The answer shows up right where you are.
Hold CtrlAlt and ask
Three examples. A draft message is selected and you say "Translate this to Spanish"; the answer replaces the draft. In an email you click the reply and say "Write a polite reply saying I can't make Thursday"; the reply is inserted. On a web page you select "HTTP 429 Too Many Requests" and ask what it means; the assistant explains it.
Or just talk it through.
A hands-free conversation that answers out loud. Cut in any time, and pick it up later.
Press CtrlAltC to start or end one
You say "Help me plan a short talk for Friday's team meeting." The assistant answers out loud: "Sure. Open with the one result you're proudest of, give two reasons it worked, and end on what's next." When the call ends, it is saved to your conversations.
Every meeting, written down.
Any call app, and no bot joins. Who said what, a summary, and the next steps.
A call starts and SpeakoFlow offers to record it. A small floating recorder shows the live transcript, split into you and the others, while the meeting is listed as recording. When it stops, the meeting is saved and the other voices are labelled Speaker 1 and Speaker 2. Notes are written with a summary and next steps for each person. Then you ask "When does the pricing page ship?" and it answers from the transcript.
Say it messy. Get it clean.
Filler words go, spoken corrections stick, and it writes in the style you pick.
You say "um so we should meet tomorrow at ten, actually no, eleven, and uh go over the budget". The filler words and the corrected time are struck out, and it types "I suggest we meet tomorrow at 11:00 to review the budget.", or a friendly, concise, or casual version of it.
Private by default.
Your voice becomes text on your own computer, with no telemetry. Nothing is uploaded unless you choose a cloud service.
On your computer
Speech, cleanup, and the assistant, offline.
- Parakeet
- Whisper
- Gemma
- Kokoro
- Voxtral
- Qwen
- Ollama
- Hugging Face
Or the cloud
Off until you add your own key.
- OpenAI
- Anthropic
- Google Gemini
- Groq
- ElevenLabs
- Mistral
- Deepgram
- DeepSeek
- xAI
- Microsoft Azure
- OpenRouter
- AWS Bedrock
- Perplexity
- Together AI
Yes. It's open source under the MIT license. There is no paid plan, no trial, no word cap, and no account.
Yes. Dictation runs on a speech model on your computer, and the assistant, AI cleanup, and meeting notes can run on the built-in engine with no internet. Cloud services are optional and use your own API key.
By default your voice is transcribed on your computer and never uploaded. There is no telemetry, no analytics, and no account. Data only leaves your computer for cloud services you set up yourself. Privacy details
No bot joins. SpeakoFlow records your microphone and your computer's audio as two streams, so it works with Zoom, Teams, Meet, Discord, Slack, or any other app. When the call ends it labels the speakers, writes a summary with decisions and action items, and lets you ask questions about what was said. On macOS, recording the other side needs a virtual audio device such as BlackHole.
Up to 99, depending on the speech model. Parakeet is the default for English, Nemotron covers 28 languages with automatic detection, and Whisper covers 99. Several models can also translate what you say into English.
No. Everything can run on models on your computer. If you prefer a cloud provider such as OpenAI, Anthropic, Gemini, Groq, ElevenLabs, or Deepgram, add your own key.
Windows, macOS (Apple silicon and Intel), and Linux (AppImage, .deb, and the AUR package speakoflow-bin). All downloads
No. SpeakoFlow isn't signed by Apple yet, and macOS shows that message for unsigned apps. Drag it into Applications, run xattr -dr com.apple.quarantine /Applications/SpeakoFlow.app in Terminal once, then open it normally. Install guide
Wispr Flow is closed source, free up to 2,000 words a week on desktop and then $15 a month, transcribes in the cloud, and has no Linux version. SpeakoFlow is free with no word cap, open source, runs on Linux too, transcribes on your computer, and also records meetings and talks back. Full comparison
You think faster than you type.
For Windows, macOS, and Linux · All downloads