A sentence can be easy to say and surprisingly tedious to type, especially when it belongs in whichever app is already open. Koi is built around that small friction: a keyboard shortcut starts dictation, and the recognized words appear at the cursor. A second shortcut brings up an AI assistant that can respond to questions about the screen and, according to the developer, perform some actions in the Mac interface.
That combination is the product's central idea. Koi is not just speech recognition, nor is it only a chat window. It tries to make voice a way to enter text and a way to ask for help with the work already in front of you. The distinction matters because those two jobs have different privacy and connectivity assumptions.
Dictation happens where you are working
The basic flow is straightforward: place the cursor in a text field, activate dictation, and speak. Koi says the text appears in the active app, including places such as mail, messaging, browsers and code editors. It also says its speech recognition runs on the Mac, so dictation can work offline and without a word limit. The app can remove filler words and format spoken lists, and the developer describes support for multiple speech models and languages.
For someone searching for a Mac dictation app, the useful detail is that Koi is intended to type into other apps rather than confining dictation to a separate editor. That makes it potentially relevant for drafting a reply, writing a note or entering prose in a code editor. The trade-off is that dictation is still an input method: it does not, by itself, understand the surrounding task or decide what the text should say.
The assistant has a different job
A separate shortcut opens Koi's assistant. You can ask about something on screen, and the app says it can point to a control, sketch an explanation, rewrite selected text or click through parts of the interface. The page describes it stopping before actions that sign in or make a payment. This is the part that turns Koi from voice to text Mac software into something closer to an assistant for the current screen.
That capability also changes the privacy picture. Koi says dictation runs locally, but its page says that questions to the AI send your words and, unless disabled, an image of the screen to the AI service. So “offline” applies to speech recognition, not to every part of the app. The page offers settings for whether the assistant sees the screen, and says it does not inspect it continuously. That distinction is worth understanding before using it around private messages, customer data or internal work.
More than the two voice tools
Koi also presents itself as a place for call recordings, transcripts and notes. The developer says it can record meetings in services such as Zoom, Meet and Teams, distinguish the local microphone from the call audio, and save the result as a document. It can also transcribe audio or video files. The page shows documents with tables, charts, tasks and study materials, alongside calendar views and an expense tracker.
This breadth may suit someone who wants dictation, meeting notes and lightweight personal organization in one Mac app. It also makes Koi harder to evaluate as a single-purpose tool. A person who only wants speech recognition may find the surrounding assistant, documents and tracking features unnecessary. The product's dock and keyboard shortcuts provide a common entry point, but the page describes a sizeable collection of functions rather than one narrowly defined workflow.
The developer says recordings and documents stay on the Mac, while AI requests are sent to an AI service. That is a useful boundary, but it is still the developer's account of data handling. Anyone considering the app for sensitive work should check the settings and the current privacy terms before relying on that distinction.
A focused fit, with clear boundaries
Koi runs on macOS and is a paid app, with a three-day free trial stated by the developer. It is not a cross-platform option, and its AI functions should not be confused with the local dictation engine. Nor does the screen-aware assistant remove the need to review its output or actions. These are meaningful constraints, not footnotes: the appeal depends on being comfortable with a Mac-only tool and with sending AI prompts, and possibly screen images, beyond the device.
If that trade-off fits, Koi brings spoken text entry and a screen-aware assistant into the same Mac workflow, with recording and documents attached. It is aimed at Mac users who regularly speak drafts or want voice help with the interface they are using. It is a poor fit for people who need Windows support, want all AI processing to stay local, or only need a minimal dictation utility.
