sideOS

Mac users who want AI help today mostly get chatbots that hand back a paragraph and leave the actual clicking, typing, and app-switching to the person. Getting things done across Mail, Calendar, Notes, and a browser means manually copying context between windows, describing what is on screen instead of just pointing at it, and running separate tools for dictation, meeting notes, and reminders. Many existing AI assistants also require sending audio, screenshots, or documents to a cloud model, which raises privacy concerns and usage costs. People taking calls, setting reminders, and connecting work tools like Slack or Notion end up stitching several disconnected apps together by hand.

sideOS sits in the Mac notch and is triggered by holding a key combo: Control+Option to give it a spoken task, Control+fn for plain dictation. It reads the live screen through macOS's accessibility tree in about 50ms, identifying real buttons and fields rather than relying on screenshots, then acts directly by opening apps, clicking, typing, pressing shortcuts, or running AppleScript across apps like Mail, Notes, Calendar, Slack, and browsers. Dictation runs on-device in up to 62 languages with no API key or cloud billing. A background task mode lets it run longer jobs such as web research, file reads/writes, PDF generation, or shell commands in a separate loop, asking approval before anything irreversible. It also records and transcribes meetings on-device, turns action items into reminders, connects to tools like Gmail, Calendar, Notion, Slack, and GitHub through MCP, and keeps an opt-in local activity log so it can answer questions about past screen activity without that data leaving the Mac.