Home / setup

Zero to local in five minutes.

Four steps. One of them is clicking a button labeled Get LLM List. By the end, you'll have asked an AI a question that never touched the internet.

1

Install Ollama — the engine

Ollama is the free, open runtime that actually executes the models. Grab the Windows installer from ollama.com and run it. It sits quietly in the background and listens only on your own machine.

# verify it's alive
C:\> ollama --version
ollama version is ✓
Privacy check: Ollama serves on 127.0.0.1:11434 — the loopback address. That's a network route that physically cannot leave your computer.
2

Pull a model — your new tenant-free brain

Models download once and are yours permanently. No API key, no per-token bill, no deprecation notice. Pick by your hardware:

Your hardwareStart withCommand
Any PC (CPU only)Small & quick (3B)ollama pull llama3.2
~8 GB VRAMThe daily driver (8B)ollama pull llama3.1
12–24 GB VRAMBigger & smarterollama pull gemma3:27b
For images (vision)llavaollama pull llava
Rule of thumb: more VRAM → bigger model → smarter answers. But even the small ones handle drafting, summarizing, and document Q&A well. You can hoard as many models as your disk allows and switch between them freely.
3

Unzip Cryptic Thinking — the cockpit

Cryptic Thinking is a portable app — there's no installer, no account, no license key. You download one .zip, extract it, and run it. The catch is that you have to extract it first: running the app straight out of the zip preview will fail, because Windows unpacks it to a temporary folder that gets wiped.

1 · Download and extract the folder

Download CrypticThinking.zip, then right-click it in File Explorer and choose Extract All… Point it at a permanent home you control. A tidy, out-of-the-way choice is your local AppData folder — paste this into the Explorer address bar to jump there, then extract into it:

# a permanent home under your user profile
%LOCALAPPDATA%
# → e.g. C:\Users\you\AppData\Local\Cryptic Thinking
Any folder works. AppData just keeps it out of sight and survives reboots. Documents\Apps\, a spot on another drive — wherever you like, as long as it's not your temporary Downloads folder and not still inside the zip. Extract the whole folder; the .exe needs the files that ship next to it.

2 · Make a Desktop shortcut

Open the folder you extracted, find CrypticThinking.exe, and give yourself a one-click launcher instead of digging through AppData every time:

  • Right-click CrypticThinking.exe → Show more options (on Windows 11) → Send to → Desktop (create shortcut).
  • A Cryptic Thinking icon appears on your desktop. Double-click it to launch from now on.
  • Want it pinned? Right-click that shortcut → Pin to Start or Pin to taskbar.
Why a shortcut? The shortcut just points back to the .exe in the folder you extracted — so keep that folder where it is. Move the folder later and you'll just need to recreate the shortcut.
Requirements: Windows 10/11. Ollama from step 1 should be running (it usually starts itself with Windows). If Windows SmartScreen warns about an unrecognized app, choose More info → Run anyway — that's the standard prompt for indie apps without a paid code-signing certificate.
4

Connect and ask

In the top-left of the app:

  • Check the address reads 127.0.0.1:11434 (the default — if you didn't change anything, it's right).
  • Click Get LLM List. Your installed models appear in the dropdown.
  • Pick one, type a question in the prompt bar at the bottom, and hit the Play button (▶) or Enter. Shift+Enter makes a line break.
That's the whole migration. Want proof it's local? Turn off your Wi-Fi and ask another question. It answers anyway. Try that with your subscription.

If something's not working

"Get LLM List" finds nothing

Ollama isn't running or the address is wrong. Open a terminal and run ollama serve, then check the address field matches 127.0.0.1:11434.

Answers are very slow

The model is likely too big for your VRAM and spilling into system RAM. Pull a smaller variant (e.g. llama3.2) — a fast small model beats a crawling big one for most work.

Image upload does nothing

Two requirements: you must be in Generate mode (API dropdown, top center — vision doesn't work in Chat mode), and the selected model must support vision (e.g. llava). The capability chips in the app show you what the current model can do.

Running Ollama on another machine?

Totally fine — point the address field at your server's IP and port instead. Your gaming PC or homelab box can serve models to a laptop running the app, all inside your own network.