Mininglamp Opens Cider+Mano-P, Turning Your Mac into a Private AI Workstation

Recently, Mininglamp has open-sourced two major local AI projects - Cider and Mano-P, which address key pain points of Mac-side inference acceleration and GUI intelligent agent operations, providing users with a complete local AI infrastructure. This means that Mac is no longer just "capable of running AI," but has truly become an efficient, private, and deeply controllable AI workstation.

Cider: Unlock the Potential of M-series Chips, Run LLM/VLM Faster and More Efficiently Locally

Many users encounter the same issue when deploying local large models on Mac: although the hardware chip performance is strong, the actual inference speed and memory usage do not meet expectations. Cider was created exactly for this purpose.

This project focuses on fully leveraging the INT8 TensorOps capabilities of M-series chips (especially M5), significantly improving the local inference speed of large language models (LLM) and vision-language models (VLM), while greatly reducing memory consumption. It provides a practical solution for Mac users to efficiently run AI models at the edge.

Mano-P: Pure Visual GUI Agent, Achieving Full AI "Screen Viewing + Operation" Process

If Cider solves the issue of "running fast," then Mano-P solves the issue of "how to use" - enabling AI to truly understand and operate computers like humans.

Mano-P is a GUI-VLA agent for edge devices, supporting local inference on Mac mini and MacBook. It breaks through the limitations of traditional agents restricted to browser operations, enabling direct control over desktop software, web interfaces, professional tools, and complex graphical workflows.

Its core capabilities include:

• Complex GUI automation operations
• Cross-system data integration
• Long-task planning and execution
• Intelligent report generation
• Self-built application development

The technical approach uses pure visual GUI operations, ensuring privacy and security as screenshots and task data remain within the device throughout the process.

Practical Demonstration: Mano-P Masters Mahjong, Achieving Autonomous Decision-Making

In the project demonstration, Mano-P has shown strong capabilities in the mahjong game scenario. It understands the game interface through pure visual perception, independently completes card identification, situation analysis, and decision-making actions, demonstrating a complete closed-loop ability from "perception" to "action."

Two Open Source Projects Working Together to Build a Local Private AI Infrastructure

Cider, as an edge inference acceleration framework, and Mano-P, as an edge GUI agent model, together form a complete local AI solution. Whether users are pursuing maximum inference efficiency or need AI to autonomously complete complex desktop tasks, they can achieve a more powerful and private AI experience on Mac.

AIbase Comment: As the demand for local large model deployment continues to grow, Mininglamp's open-sourcing of Cider and Mano-P precisely addresses the pain points of Mac users, providing important references for the practicalization and ecosystem development of edge AI. Interested developers can follow the latest project updates and build their own local AI workstations.