Buddy

Buddy is a voice assistant for Mac that can answer questions about what is visible on the screen, explain error messages, point to on-screen elements, draft replies from a thread, and handle dictation. It is built to work entirely on the user’s Mac and is presented as a privacy-focused assistant for situations where screen content should not be sent elsewhere.

Its interaction model is push-to-talk: the user presses a key, speaks, and Buddy captures a single snapshot of the screen at that moment. The assistant then reads the screen with Apple’s Vision framework, transcribes speech with a local speech model on the Neural Engine, and uses an on-device language model to produce answers and identify the element it means. When it knows the target, it grounds that to coordinates and moves the cursor to it; if it is unsure, it does not point. The page also says it can provide step-by-step walkthroughs and that dictation works by holding the key and speaking into any text field.

Buddy is described as running 100% on the user’s Mac with no account, no API key, and no server. It also states that there is no background watching, that the snapshot is discarded after the answer, and that update checks are the only network traffic mentioned. The maker is identified as LocalWispr. It is offered through a waitlist, with waitlist members said to get it first at founder pricing, and the site says it is a one-time purchase with no subscription and a final price at launch. The required platform is an Apple Silicon Mac, M1 or newer, and the newest macOS is said to give the best answering model.

The page frames Buddy as a tool for people who cannot use cloud AI at work or who handle privileged, regulated, or confidential material on screen. It also says it will not write an app from a screenshot, browse the web, or do deep research, and that if it cannot see the answer on the screen it says so rather than inventing one.