Tap Fn on macOS or Right Alt on Windows in any app, speak naturally, and polished text appears where your cursor is. AI cleanup fixes grammar and self-corrections, and Pure Offline Mode keeps everything on your device.
Click and hold the keycap below (or hold Space on this simulator). Hold Shift as well to switch to Action Mode.
Just Talk is free, open-source software built independently without expensive corporate signing certificates ($99/yr Apple Developer / $400/yr Windows EV). Here is how to complete first-time setup safely in 10 seconds:
JustTalk-macOS.dmg and drag Just Talk to your /Applications folder.xattr -cr /Applications/JustTalk.app
JustTalk-Windows.exe installer.Choose the speech engine that fits you — Apple and Windows built-in recognition, Google, or Whisper running entirely on your hardware — with text typed straight into the app you're using.
With Pure Offline Mode, CTranslate2 int8 Whisper runs locally on ARM NEON and AVX-512. Audio stays in memory and is discarded immediately after decoding — nothing is sent over the network.
Select text in any app, hold your trigger key + Shift, and say “make this shorter” or “translate to Nepali” — the selection is rewritten in place. With nothing selected, Action Mode writes what you ask for — “write me a sick leave email” or “write a short poem about rain” — right where your cursor is (needs AI formatting turned on).
Only download models for the languages you actually speak. If you only speak English, download the English model (~460 MB) to save disk space and RAM.
Types text into whatever app is currently active (VS Code, Slack, Notion, browsers, terminals). Your clipboard is restored automatically, long dictations are kept whole across pauses, and Ctrl+Cmd+V (Win+Alt+V on Windows) pastes your last dictation again.
Total freedom over your AI stack. Connect local Ollama (Llama 3.2, Qwen 2.5, DeepSeek R1) for 100% offline, private text cleanup with zero token costs. Or load any custom Hugging Face repo or local CTranslate2 speech model with live latency benchmarks.
One toggle in Settings completely disables all external API calls. Your speech is transcribed locally and pasted with zero network connections.
Background noise reduction and WeSpeaker CAM++ speaker recognition are available in Settings › Voice Isolation, so Just Talk can tell your voice from other people's. Noise reduction is light-touch on purpose: heavy denoising tends to make speech recognition less accurate.
Whisper Large-V3 struggles with Nepali, especially conversational speech. Just Talk can now use Kriti, an open Nepali speech model by Naamche Labs (119M parameters, MIT licensed), built on AI4Bharat's Nepali IndicConformer. It runs fully offline: your voice never leaves your computer.
Combined with optional AI formatting, mixed Nepglish (e.g. "yo code ma error aayo, fix garnu paryo") is cleaned and formatted into fluent Devanagari, Romanized Nepali or English.
Naamche Labs' published result — the conditions closest to dictating into a microphone. Runs fully offline, with no account or API key.
| Feature | Just Talk (Open Source) | Cloud Voice Typing Apps | Built-in OS Dictation |
|---|---|---|---|
| On-Device Audio Privacy | Pure Offline Mode: 0 audio sent | Streams audio to the cloud | Partial (device dependent) |
| Edit selected text by voice (Fn + Shift) | Rewrites the selection in any app | Requires clicking dropdowns | Unsupported |
| Conversational Nepali & Nepglish | Nepglish, Devanagari & English output | High error rates on slang and code-switching | Unsupported |
| Pure Offline Operation | One toggle disables all network calls | Requires an internet connection | Basic language models only |
| Pricing & Licensing | 100% Free & Open Source (MIT) | $10 - $20 / month subscription | Bundled (Basic accuracy) |
Your microphone is only recorded while dictation is active. Audio is kept in memory and cleared after transcription — it is never saved to disk. Whether audio is sent to a speech service depends on the engine you pick; Pure Offline Mode keeps it on your computer.
API keys for AI formatting are stored only on your computer, in Just Talk's private app-data folder with owner-only permissions, and are sent only to the provider you selected. AI formatting sends text, never audio.
If you handle sensitive medical records, proprietary code, or confidential legal documents, toggle Pure Offline Mode. Just Talk bypasses every cloud service and types on-device transcriptions straight into your app.
Yes. Just Talk is free and open source under the MIT license. There is no subscription, no account and no usage limit. If you turn on AI formatting with a paid provider, you pay that provider directly for the text it processes.
It depends on the speech engine you choose. With Pure Offline Mode, or the on-device Whisper engine, audio never leaves your computer. The built-in Apple engine uses on-device recognition when your Mac supports it for your language. The Google Web Speech engine — used for Nepali, and first on Windows when you are online — sends audio to Google to transcribe it. AI formatting only ever sends text, never audio.
Tap Fn on macOS or Right Alt on Windows to start, speak, then tap again to type the text into the app you are using. Prefer holding the key while you talk? Turn on “Hold the shortcut while speaking” in Settings › Shortcut. Press Esc to cancel.
Yes. Just Talk records for up to 5 minutes and keeps everything you said across pauses. In tap mode it stops on its own after 8 seconds of silence, so you can stop to think without losing your place.
It adds punctuation, fixes grammar and spelling, removes filler words like “um” and “you know”, and resolves self-corrections. Say “can we meet at five, actually four” and Just Talk types “Can we meet at four?”. It never summarizes or changes what you meant, and you can turn it off with one toggle.
Google Gemini, OpenAI, Anthropic Claude, xAI Grok, Groq, OpenRouter, DeepSeek, any OpenAI-compatible API (for example NVIDIA), or Ollama running locally on your computer for completely private, free cleanup. Bring your own API key — it stays on your device.
Yes. Turn on Pure Offline Mode in Settings › Voice & Speech. Speech is transcribed by on-device models and nothing is sent over the network. Offline speech needs a one-time model download, sized to the languages you choose.
Tick the languages you speak once in Settings › Languages. Then choose what you are typing in right now with the Typing in buttons on the Home screen (English, नेपाली or Mixed) or from the tray / menu bar icon. Turn on Translate to English if you want any language typed in English.
Yes. Speak Nepali, English or a mix of both. Nepali can be recognised fully offline by Kriti, the open Nepali speech model by Naamche Labs. Just Talk can type Romanized Nepali (“k cha”), Devanagari (“के छ”), or translate to English — and in Automatic mode it picks the right one for each app, such as Romanized in chat apps and Devanagari in documents.
Any app with a text field: VS Code, Cursor, Slack, Chrome, Word, Notion, Gmail, WhatsApp, terminals and more. If no text field is focused, the text is copied to your clipboard so you can paste it yourself.
Yes. Choose “Translate to English” on the Home screen. Whatever language you speak is typed as fluent English.
Press Shift while dictating to switch to Action Mode. Instead of typing your words verbatim, Just Talk follows them as an instruction — for example translating them into English.
Yes. Select the text in any app, hold your shortcut plus Shift (Fn + Shift on macOS, Right Alt + Shift on Windows) and say what to change, such as “make this shorter”, “make it more polite” or “translate to Nepali”. Just Talk rewrites the selection in place. This needs AI formatting turned on.
Press Ctrl + Cmd + V on macOS or Win + Alt + V on Windows. Just Talk types your most recent dictation into the app you are using now — useful when the cursor was in the wrong place. You can also choose Copy Last Dictation from the tray or menu bar icon.
Just Talk is an independent open-source project and isn't signed with a paid Apple or Microsoft certificate. On macOS, run xattr -cr /Applications/JustTalk.app once in Terminal (or right-click the app and choose Open). On Windows, click “More info” then “Run anyway”. All source code is public on GitHub.
macOS 12 or later on Apple Silicon or Intel, or Windows 10/11 (64-bit), plus a microphone. On-device models need 0.5–1.5 GB of disk space, depending on the languages you choose.
Open Settings › Updates & Startup and click “Check for updates”. When a new version is available, Just Talk downloads and installs it in place and keeps all your settings, history and API keys.
Only on your computer, in Just Talk's private app-data folder (~/Library/Application Support/JustTalk on macOS, %APPDATA%\JustTalk on Windows). Keys are saved with owner-only permissions and sent only to the provider you selected.
Yes. Pick your languages during setup or in Settings › Languages & Models, and Just Talk downloads only the models those languages need — about 460 MB if you only speak English.
Yes. Point Just Talk at any faster-whisper model on Hugging Face or a local folder for speech recognition, and use Ollama (for example Llama, Qwen or DeepSeek) for AI formatting with no internet and no token costs.
It is free and open source, works on both macOS and Windows, lets you choose — or run locally — the AI that cleans up your text, has a fully offline mode, and has first-class Nepali and Nepglish support.