Think out loud.
I’ll write it right.
— your ThinkInk
ThinkInk listens to your voice and types finished text into any Windows app: no “um”s, correct punctuation, a professional tone or another language — whatever you ask for.
You speak like a human.
The text comes out like it’s been edited.
- Click the microphone in the widget or press your shortcut, and speak into any field: email, messenger, document.
- Hesitations disappear. “Um”s, repeats and “no, wait” are removed, and the meaning stays yours.
- Finished text appears right where your cursor is. No manual copying.
Example: dictated “I’ll, um, send you the estimate tomorrow, no, on Friday. I still need, need to check the numbers. The total might change a bit.” Result: “I’ll send you the estimate on Friday. I still need to check the numbers. The total might change a bit.”
One thought — eleven ways to put it
Each mode has its own color, and each can have its own shortcut. Pick a mode in the widget, and the same dictated phrase becomes an email, a prompt or a translation.
Correction
Examples show how each mode works. The actual result depends on the model you choose and how you speak.
Select a paragraph. Say what to do with it.
“Make it formal”, “translate into Spanish”, “shorten it but keep all the dates”: in plain words, no commands to memorize. ThinkInk reworks the passage and copies the result to the clipboard. You paste it with Ctrl + V wherever you need it.
The clipboard works in every app, so the result isn’t tied to the window the text came from. Paste it over the selection, into another document, an email or a chat. And you can say the command no matter which window is open.
A widget that stays out of your way
A glass capsule floats above your windows and lets clicks through where it’s transparent. Drag it anywhere, shrink it to a thin strip or expand it to show the mode name.
Full widget: mode, name and record button in one capsule.
- Two themes — Crystal and GraphiteLight, dark or matching the system. Widget and menu glass are set separately.
- “Result copied”After editing, the text is already on the clipboard, and a small card reminds you: Ctrl + V.
- History with undoSearch, filter by mode and restore your last ten deletions.
- Statistics and costsHow many words you dictated in 7, 30 and 365 days, plus a cost calendar for each service.
- Processing without complicated setupYou choose who recognizes your voice and who works on the text. Anything missing shows up right away, and a short seven-step guide tells you what to do.
- Tutorial on the real widgetA built-in walkthrough takes you through dictation, translation and editing step by step.
Voice recordings take up no disk space
Audio exists only in memory while it’s being processed, and is never saved as a recording.
History and settings stay in your Windows profile. There’s no separate server holding your texts.
API keys are encrypted by Windows and never shown in the interface.
Recognition can work offline. Whisper runs on your computer, with or without an NVIDIA graphics card. You can choose the model: from the fast Small to the accurate Large v3.
If you choose OpenAI or Gemini for voice, the audio is sent there. For text work (modes, translation, editing), ThinkInk sends text only to the service you chose: OpenAI, Claude or Gemini. The free Gemini key has its own terms: Google may use the texts you send to improve its service.
What you get
ThinkInk’s main features and what each one gives you.
| Feature | What it gives you |
|---|---|
| Free app | ThinkInk itself costs nothing. Text processing runs through your own subscription or API key — you only pay the service, if it charges. |
| The AI of your choice | For text — OpenAI, Claude or Gemini: through an account (ChatGPT, Claude Code) or an API key. For voice — Whisper on your computer, OpenAI or Gemini. All keys go in one place, and ThinkInk figures out which service each key belongs to. |
| Recognition on your computer | Whisper turns voice into text offline — with or without an NVIDIA graphics card. |
| 11 ready-made modes | Correction, professional and friendly tone, email, prompt, “concise” and more. |
| Editing by voice | Select some text and say what to do with it: “make it formal”, “shorten it but keep the dates”. The result goes straight to the clipboard, so you can paste it anywhere. |
| Translation while you dictate | You speak one language — text in another appears in the field. |
| Any Windows app | Text is inserted where the cursor is: in email, a messenger, a document. Nothing to copy. |
| Your voice isn’t stored | Audio exists only in memory, history stays in your Windows profile, API keys are encrypted. |
| Seven interface languages | The interface is available in English, Ukrainian, German, Polish, Spanish, French and Portuguese. |
| History and statistics | Search, restore deleted entries, see how many words you dictated and what processing cost. |
Good to know up front
- This is a beta (β) version — Windows x64 only.
- The app is free for now. A subscription or API key billing is separate, with the service itself.
- Claude with a subscription works through Claude Code installed on your computer and responds more slowly than with an Anthropic key. Gemini connects only with an API key.
- Exact OpenAI and Anthropic costs are shown with an admin key. For Gemini, or without such a key, ThinkInk shows an estimate.
- Live dictation works with Whisper or OpenAI, not with Gemini.
- Some apps running with elevated rights don’t accept automatic pasting — then you can copy the text from the history.
- One recording — up to 15 minutes.
Frequently asked questions
How to set it up, what it costs and what computer you need.
1How does ThinkInk work?
In two steps. You pick a service for each step.
What you can use for each step
Free speech recognition right on your computer, no internet needed. ThinkInk downloads it for you.
- Voice → transcript
- Transcript → finished text
Like a password for an AI service: OpenAI, Anthropic (Claude) or Gemini. You create it on the service’s website in a few minutes. Marked below as: API
- Voice → transcript (OpenAI, Gemini)
- Transcript → finished text
Your paid ChatGPT or Claude account. You just sign in, no key needed. Marked below as: account
- Voice → transcript
- Transcript → finished text
2What should I choose?
Why: in tests, GPT-6.1 Terra handled every mode accurately and never answered a question instead of writing it down. Claude Sonnet 5.5 also did well. Processing costs about $0.002–0.004 per phrase.
- Voice → transcript
- Whisper Large v3orOpenAI API
- Transcript → text
- Claude Opus 5.5 APIorOpenAI GPT-6.1 Sol API
≈ $0.004–0.007 per phrase
≈ 150–300 phrases per $1. Any other strong model with a key works well too.
- Voice → transcript
- Whisper with NVIDIAorOpenAI API
- Transcript → text
- Any model API
- Voice → transcript
- Whisper on your computer
- Transcript → text
- OpenAI GPT-6 Luna API
≈ $0.0002 per phrase
≈ 5,000 phrases per $1
- Voice → transcript
- Whisper on your computerorGemini API (free)
- Transcript → text
- Gemini API (free)
$0
Free Gemini terms are in question 5.
Recommendations and speeds come from testing: more than 150 text-processing checks (the same phrases across different modes and models). Prices are estimates based on service rates as of November 1, 2026.
3How much does it cost?
ThinkInk itself is free. You only pay the AI service, and not always.
0¢
- Voice → transcript
- Whisper on your computerorGemini API (free)
- Transcript → text
- Gemini API (free)
Just dictation, no processing? Whisper is enough, no key needed.
No extra cost
- Voice → transcript
- Whisper on your computer
- Transcript → text
- ChatGPT accountorClaude Code account
Within your plan’s limits. Claude requires the Claude Code app.
< 1¢ per phrase
- Voice → transcript
- Whisper on your computerorOpenAI API
- Transcript → text
- OpenAI APIorClaude APIorGemini API
You only pay for what you use. Voice through OpenAI costs up to 0.6¢ per minute. You can see spending in the “Costs” section.
Prices are estimates based on service rates as of November 1, 2026.
4Why is ThinkInk free?
I built ThinkInk for myself, my family and my friends. I polished every detail and tested it in every way it can be used. I think it turned out really handy, so I want to share it with everyone. I’ll be glad if it helps you in your daily work or conversations.
If you like ThinkInk, I’d be grateful for your feedback. And if you can — for your support too.
- Right now the app is completely free
- Feedback and support help make it better
- Paid modes or features may appear later
5How do I use Gemini for free?
All you need is a free Gemini key. A Google AI subscription won’t work: Google doesn’t allow it in other apps.
- Step 1Open Google AI Studio and sign in with your Google account.
- Step 2Create a key. Don’t turn on billing.
- Step 3Paste the key into ThinkInk under “API keys”.
What to know about free Gemini
- Google may use your texts, and people may read them.
- In the EU, the UK and Switzerland, the paid version is required.
- There’s a daily request limit.
- Live dictation doesn’t work with Gemini.
6What computer do I need?
- System
- Windows 64-bit
- Graphics card
- Fastest with NVIDIA. Works without it too, just slower.
- Memory
- From 1 GB free (Small) to 4.7 GB (Large v3)
- Disk space
- 0.5 to 3.1 GB, depending on the model
- Internet
- Only to download the model
- System
- Windows 64-bit
- Graphics card
- Not needed
- Memory
- None needed for a model
- Disk space
- No model needed
- Internet
- Needed at all times
Memory figures are approximate, from the developers of Whisper for PCs (whisper.cpp).