Your AI that actually sees and does.
Brah is a desktop voice assistant that lives in the corner of your screen. Talk to it and it listens, looks at your screen, controls your computer, and gets things done, all in realtime through the OpenAI Realtime API.
It's not just a chatbot. It takes screenshots and reasons about what's on screen, drives your mouse and keyboard, automates the browser, and manages your tasks and calendar, hands-free.
Most voice assistants can talk but can't do. Brah closes that gap: it sees your actual screen, controls your actual computer, and runs real tools, so a conversation turns into action instead of just suggestions.
Low-latency voice in, voice out, powered by the OpenAI Realtime API. Live transcription and natural barge-in so you can interrupt and steer mid-sentence.
Take screenshots of any window or display and have the model analyze what's there: visible text, UI elements, errors, and the obvious next action.
Two modes:
- Browser mode: A sandboxed Chromium browser (via Playwright) for web tasks
- OS mode: Full OS-level control (via nut.js) to drive the real mouse and keyboard
Built-in tasks and calendar the agent can create, update, and remind you about.
Pulls live information and reads pages on demand.
git clone https://github.com/KenKaiii/brah.git
cd brah
npm install
npm startSign in to OpenAI from inside the app to start a Realtime session. That's it.
- Data (planner, activity, screenshots) is stored locally on your machine
- Credentials are encrypted via your system keychain (Electron
safeStorage) - No analytics, no telemetry
npm run check # format + lint (Biome)
npm test # check + Node test suite
npm run build:macStack: Electron + OpenAI Realtime API + Playwright + nut.js
Code signing is auto-discovered from your keychain (or CSC_* env vars); with no certificate the macOS build falls back to ad-hoc / unsigned.
- YouTube @kenkaidoesai - tutorials and demos
- Skool community - come hang out
MIT
An AI that sees your screen, runs your computer, and gets things done, hands-free.