A personal AI agent that lives on it's own Mac, chats like humans through your daily chat apps, works like a human through commands, browser automation and desktop control.
Real conversations. Real tasks. Fully autonomous.
An always-on agent with eyes and hands on computer.
Sees screen just like a human — describes UI elements, reads images, and interprets documents to decide what to do next.
Controls mouse and keyboard, operate any desktop apps, browsers, social apps, etc. and runs terminal commands.
Connects to any MCP servers and installs Skills on the fly to expand its capabilities without limits.
Has its own Gmail account for email and Google Calendar for scheduling recurring tasks.
Can learn new skills, create/install them at runtime, and upgrade itself to the latest version.
Remembers context across sessions with SQLite-backed memory, so it never forgets your preferences or past tasks.
Node.js >= 24.12.0, pnpm, macOS, an Anthropic API key, and a dedicated Gmail account for the agent.
Clone the repo, install dependencies, and build.
The first run starts an interactive onboarding wizard that configures your API keys and channels.
git clone https://github.com/babelcloud/visionclaw.git
cd visionclaw
pnpm install
pnpm build
pnpm start