Open-source voice AI · runs on your device
Speak
It's written
Dictate into any app, take live meeting notes, and turn text into natural speech. Free, unlimited, and private, because it all runs on your own machine.
Use it from
Download the app, then press Ctrl+Shift+Space in any app and talk.
DictationNotetakerText to speech99 languagesOfflineNo uploadsMIT licensed
Everything voice
One workspace for your voice, in and out
Dictation
Talk and get clean text. Fillers vanish, “scratch that” works, snippets expand, and nothing is uploaded.
Open →02Notetaker
Live transcripts of your calls with you and the others labelled, then highlights, actions and decisions.
Open →03Speech studio
Turn anything from a line to a novel into natural speech, with playback that starts instantly.
Open →04Transcribe files
Audio and video to text and subtitles in 99 languages, or translated into English.
Open →05Long-form projects
Multi-voice audiobooks and podcasts, mastered to a single file.
Open →0628 voices
American and British narrators, from warm storytellers to crisp presenters.
Open →Integrations
Talk to the tools you already use
Google Docs
Chrome extension: dictate into Docs, Gmail, Notion and any text box on the web.
Microsoft Word
An add-in panel that writes at your cursor and reads your document aloud.
Claude Code
Speak your prompts into the terminal, and let Claude talk back through Kelados.
Every other app
The desktop app types wherever your cursor is: Slack, VS Code, email, anything.
Text to speech
Give any text a voice
28 natural voices generated in your browser. No credits and no character caps. Try it right here.
- Characters spoken
- —
- Minutes of audio
- —
- Generations
- —
- Members
- —
The first run downloads the model once (~90 MB+)
How it's free
They charge for their GPUs. Kelados uses yours.
Download once
Your device fetches compact open models (Moonshine, Whisper, Kokoro) from a public CDN and caches them.
Run locally
WebGPU, or multi-threaded WASM on older machines, runs them faster than real time.
Keep it private
Your voice and text never leave your machine. The optional account only syncs settings.
| Kelados | Paid voice apps | |
|---|---|---|
| Price | Free, forever | Monthly subscription |
| Word or minute limits | None | Quotas and tiers |
| Where speech is processed | Your device | Vendor servers |
| Your audio is uploaded | Never | Every time |
| Works offline | Yes, after first load | No |
| Source code | Open (MIT) | Closed |
Compared with typical paid services. Check each vendor for current terms.
For developers
Three lines. No invoice.
Use the Node SDK, the CLI, an MCP server for AI agents, or point an existing TTS client at a server you run yourself.
Read the docs →// npm install kelados import { createKelados } from "kelados"; const tts = await createKelados(); const clip = await tts.speak("Hello, world", { voice: "af_heart" }); await clip.save("hello.wav");
FAQ
Questions, answered
What is Kelados?
+
Kelados is a free, open-source voice AI app. It does voice dictation into any app, live meeting transcription with notes, file transcription in 99 languages, and natural text to speech with 28 voices. All speech processing runs on your own device.
Is Kelados really free?
+
Yes. There are no word, minute or character limits and no subscription. Because the AI models run on your computer instead of a server, there is no usage cost to charge for. The code is MIT licensed.
Does Kelados upload my voice or audio?
+
No. Speech recognition (Moonshine and Whisper) and speech generation (Kokoro) run locally in your browser or the desktop app. Your audio never leaves your device; an account is only needed to sync projects and settings.
What is the best free alternative to Wispr Flow?
+
Kelados offers the same core workflow as Wispr Flow (press a shortcut, talk, and clean text appears in any app, with filler words removed and spoken corrections understood) for free, with recognition on your device.
Can Kelados take meeting notes without a bot joining the call?
+
Yes. The Kelados notetaker records your microphone and the audio of the meeting tab you share, so nothing joins your Zoom, Google Meet or Teams call. You get a live transcript, then highlights, action items and decisions.
Which languages does Kelados support?
+
Live dictation is tuned for English. File transcription supports 99 languages and can translate them into English. Text to speech has American and British English voices.
Does Kelados work offline?
+
Yes, after the first use. Models download once (about 100 MB for dictation) and are stored on your device, so recognition and speech generation work without an internet connection.
Can I use Kelados with Claude Code?
+
Yes. Use the desktop app to speak prompts into the terminal, and add the Kelados MCP server (claude mcp add kelados -- npx -y kelados mcp) so Claude Code can talk back.