See it in action
Twenty-six seconds, in one take. A question, a Wikipedia lookup it decided to make, and an answer, all on the Mac you are looking at.
What people actually use it for
It answers much like the assistants you have used in a browser. The difference is where the answer is computed.
-
Ask anything
Untangle a lease clause, plan a week of dinners, check the math on a loan.
-
Drop in a document
A PDF, a Word file or a spreadsheet. Ask which line grew fastest and it reads the table to answer.
-
Show it a picture
A photo of a whiteboard, a menu or a street sign. Some models also listen to audio and watch video.
-
Watch it think
The stronger models reason before they answer. Read it, cut it short, or switch thinking off for speed.
-
Let it look things up
Wikipedia, today's headlines, the weather and web search, each behind its own switch.
-
Talk instead of typing
Tap the microphone and speak, and have replies read aloud in one of eight voices.
It looks things up.
Ask about something it cannot know and it searches Wikipedia or the web, shows you each step, and answers from what it read.
Hand it a PDF.
Drop in a document and ask about the table inside it. It reads the file itself: the answer here came out of the four rows in that report, not out of the filename.
Show it a picture.
Attach a photo and ask what is in it, what the sign says, or what to cook from what is in the fridge.
Where your words go
Nowhere, unless it looks something up.
Most assistants send your sentence to a company. ChatᵒᵏF has nowhere to send it: the whole model sits in your Mac's memory and runs on its own chip. Privacy here is not a policy that could be revised next year. It is the shape of the thing.
Wikipedia and web search are the one time anything leaves. For Wikipedia the matching happens on your Mac first, so what travels is the article it decided to open, not the question you asked. For the web, the search terms the model writes go to a search service. One button, airplane mode, switches both off.
No account. No sign-in. No subscription. No ads, no analytics, no telemetry.
It remembers, in notes you can read
Tell it to remember something and it writes a short note, and the chat says what it wrote. Later questions are checked against your notes first, whichever model is answering.
Every note is listed under Memories, to read or to delete. They are plain text files on your Mac, so switching models keeps everything the app has learned. One switch, Total Recall, turns all of it off.
Pick a model, or keep the one it starts with
Several open models are offered, and the list shows only the ones your Mac can actually hold, so every one you see will run. A small one answers almost instantly; a large one is noticeably smarter and wants a Mac with memory to spare.
Each model is downloaded once, its integrity checked, and then works offline. On a MacBook Air, Gemma 4 E4B writes about 27 words a second; the numbers for every model are on GitHub.
Prefer the command line?
Homebrew installs the same signed and notarised build.
brew tap leok7v/tap
brew trust leok7v/tap
brew install --cask chatokfThe middle line is Homebrew's own gate on taps that are
not its own; it asks for it by name if you skip it. The cask lives in a
separate repository because brew tap clones a whole repo,
and this one carries a 132 MB model file.