Mano-P Lets Mac Users Run a Local GUI Agent Without Cloud or API Keys

Mano-P is an open-source GUI agent for Apple Silicon Macs that automates screen interactions — such as clicking buttons and filling forms — using a vision-language model running entirely on-device. The agent operates in a loop: it captures a screenshot, reasons about the next action, executes it, and repeats until the task is done. It uses the Cider SDK for MLX 8-bit quantization, which speeds up inference by quantizing both model weights and activations, unlike most tools that only quantize weights. Installation requires an Apple M4 chip or newer, at least 32GB of RAM, and macOS permissions for Screen Recording and Accessibility. The setup is managed via a CLI tool called mano-cua, installable through Homebrew, and the quantized 4B model uses roughly 4.3GB of memory at peak.