Get started with Wink

Requirements

1Install Ollama

Ollama is a free app that runs AI models on your own computer. Wink uses it as its “brain”, which is why your screen never has to leave your PC.

1. Go to ollama.com/download/windows and click Download for Windows.
2. Run OllamaSetup.exe and click Install.
3. That’s all. Ollama runs quietly in the background (you’ll see a small llama icon near the clock). You don’t need to open it or type any commands.

2Install Wink

1. Download the Wink installer (Winky-AI-Setup-x.y.z.exe).
2. Open the file. Choose where to install it, or keep the default, and click Install.
3. Leave Run Wink AI ticked and click Finish.

3First start: choose your model

When Wink starts for the first time, it opens a Set up Wink window:

1. It checks your graphics card. You’ll see its name, memory, driver and CUDA version. A green dot means Wink will run on your graphics card. A yellow dot explains what to do, for example “Update the driver” if it supports a CUDA version older than 12.
2. Pick a model. Wink marks the one that suits your PC as Recommended for your PC: Fast (3B), about 3 GB, or Accurate (7B), about 6 GB. See Choosing a model.
3. Click Download. A progress bar shows how far it is. You can keep using your PC meanwhile.
4. Click Launch Wink now. Wink loads the model and checks that it runs on your graphics card. Then the window closes and a notification says “Wink is ready”.

After that, Wink lives in the system tray: look for the blue pointer icon near the clock. If you don’t see it, click the ^ arrow to show hidden icons. Wink also downloads its voice models once (about 400 MB).

If Launch says the model runs only partly on the graphics card, Wink still works but answers are slower. The Fast (3B) model usually fits fully.

Using Wink

Normal shortcuts like Ctrl + C never start voice. Voice only starts when you hold Ctrl on its own for a moment. A live waveform shows Wink can hear you.

Guide me vs. do it for me

Questions (“how do I…?”) are guided:

• A ring points at exactly what to click, with a checklist of every step.
• Wink waits for your click, then points at the next thing.
• Press ▶ Do it for me, or say “do it”, to let Wink finish the task itself.

Commands (“open Word and write…”) are done for you:

• Wink shows its plan, then carries it out step by step.
• Every action has a short countdown first, and Wink checks the screen after each step.
• For writing, you’ll see “Writing… 120 words” while it works, then it types the text in.

To have Wink always do tasks itself, right-click the tray icon and choose Wink clicks for me.

Things to ask

Short, clear requests work best. For long tasks, it helps to ask one part at a time.

The tray menu

Right-click the blue pointer icon near the clock:

Choosing a model

You pick the model in the setup window, and Wink recommends one based on your graphics card’s memory. Both are free.

Without an NVIDIA card, both run on the processor. That works, but each step can take 20 seconds or more.

To switch later, right-click the tray icon → Choose AI model…, pick the other model, and click Download or Launch Wink now.

Settings

Right-click the tray icon → Open settings file, change what you need, save, then choose Reload settings.

{
  "taskMode": "guide",
  "voiceAutoSend": false,
  "ollamaModel": "qwen2.5vl:3b",
  "kokoroVoice": "af_heart",
  "speechSpeed": 1.05,
  "speak": true
}

Troubleshooting

“I need Ollama to think”

Ollama isn’t installed or isn’t running. Install it from ollama.com. If it’s installed, start Ollama from the Start menu. Wink picks it up within a few seconds.

“I’m still downloading my AI model”

The model hasn’t finished downloading. Right-click the tray icon → Choose AI model… to see the progress. If it says Something went wrong, check your internet connection and free disk space, then click Try again.

“It runs on the processor” or “only partly on the graphics card”

Wink checks this when you click Launch Wink now. Update your NVIDIA driver from nvidia.com/drivers if the setup window asks you to, and choose the Fast (3B) model on cards with less than 10 GB.

“I got confused there”

The model gave an answer Wink couldn’t read. Ask again, a little more simply. If it keeps happening, right-click the tray icon → Open log file and send the file to support.

“I didn’t catch that”

Wink didn’t hear any speech. Check that your microphone is on and selected in Windows (Settings → System → Sound → Input), then hold Ctrl while you speak. You can always type with Ctrl + Alt.

It’s slow

• The first request after starting takes 15–30 seconds while the model loads. Later ones are much faster.
• Without an NVIDIA graphics card, each step takes longer. Closing games and other heavy apps helps.
• If you picked Accurate (7B) on a card with less than 10 GB, switch to Fast (3B) with Choose AI model… in the tray menu.

It clicked the wrong thing

Press Esc or move the mouse to take over. Asking again with a more specific request (“click the blue Save button at the top”) usually helps. A bigger model is also more accurate; see Choosing a model.

It can’t control an app

Windows doesn’t allow normal apps to control windows that run as administrator. Wink can still guide you there; you just do the clicking.

I can’t find Wink

Look for the blue pointer near the clock, or click the ^ arrow to show hidden icons. If it isn’t there, start Wink AI from the Start menu.

Uninstall

1. Open Settings → Apps → Installed apps, find Wink AI, and click Uninstall.
2. Your settings and voice models are kept, in case you reinstall. To remove them too, first choose Open data folder from Wink’s tray menu and delete that folder.
3. The AI model belongs to Ollama. To free that space, uninstall Ollama the same way, or run ollama rm qwen2.5vl:3b (or qwen2.5vl:7b) in a terminal.

Privacy

• Wink’s AI, speech recognition and voice all run on your PC.
• Screenshots go from Wink to Ollama on the same computer and are never uploaded.
• The internet is used only to download the models during setup.
• There’s no account, no sign-in and no tracking in the app.