Mano-P
Mano-P is an open-source GUI-VLA (vision-language-action) agent designed for edge devices. It enables pure vision-driven, cross-platform GUI automation through natural language instructions. The name Mano means hand, with P standing for Private, reflecting its focus on local-first AI. Mano-P currently ranks first on the OSWorld benchmark in the specialized category with a score of 58.2%. It runs locally on Apple M4 Mac mini or MacBook devices with 32GB or more of RAM, or via a USB 4.0 compute stick, ensuring all data stays on-device with no information leaving the user hardware. The project supports complex multi-step task planning and execution, making it suitable for automating GUI workflows across diverse applications. It is released in three progressive phases: Mano-CUA Skills for agent enthusiasts building intelligent computer-use workflows, local models and SDK for developers needing secure on-device inference, and training methodologies with pruning and quantization techniques for those who want to bui