NumCraft: Natural Language-Driven CNC Machining with Multi-Agent Collaboration for Safe G-Code Generation

NumCraft is an open-source innovative project addressing the high technical barrier of traditional CNC programming. It uses multi-agent LangGraph workflows to convert natural language descriptions into safe NC G-code, automating CNC machining process planning, tool selection, path optimization, and other links, providing a convenient programming solution for the intelligent manufacturing field.

Computer Numerical Control (CNC) machining is a core technology in modern manufacturing, but traditional CNC programming requires operators to master professional G-code and machining process knowledge, which has a high threshold. With the development of artificial intelligence, simplifying CNC programming using natural language interaction has become an important research direction in intelligent manufacturing. NumCraft is a solution developed for this demand, combining large language models and multi-agent architecture to realize automatic conversion from natural language to safe G-code.

NumCraft uses LangGraph to build multi-agent workflows, where each agent collaborates with clear division of labor: the requirement parsing agent extracts key parameters such as workpiece material and machining type; the process planning agent formulates machining sequence, strategy, and cutting parameters; the tool selection agent recommends optimal tool combinations; the path planning agent generates efficient motion paths; the safety verification agent reviews G-code safety (overtravel, collision, etc.). The LangGraph workflow supports conditional branching, loop iteration, parallel execution, and state management, improving collaboration efficiency.

NumCraft has practical value in multiple scenarios: in rapid prototyping, it shortens programming time and quickly verifies design manufacturability; in small-batch custom production, it flexibly adapts to design changes and reduces reliance on professional programming; in education and training, it helps students understand the relationship between G-code and machining instructions, lowering the learning threshold; in process knowledge inheritance, it captures expert experience, standardizes best practices, and supports team collaboration and sharing.

The core technical highlights of NumCraft include: safety-first design (multi-level verification, conservative default strategy, boundary condition handling, simulation verification); domain knowledge integration (material cutting characteristic database, tool parameter database, etc.); scalable architecture (supports multiple CNC control systems, integrates CAD/CAM, custom post-processors, and open plugin interfaces).

Comparison between NumCraft and traditional CAM software: In terms of user interface, traditional CAM uses graphical operations while NumCraft uses natural language interaction; in learning curve, traditional CAM requires professional training while NumCraft is intuitive and easy to understand; in programming efficiency, traditional CAM relies on operator experience while NumCraft uses AI-assisted automation; in flexibility, traditional CAM is limited by software functions while NumCraft is open and scalable; in cost, traditional CAM requires commercial license fees while NumCraft is open-source and free. NumCraft does not replace traditional CAM but provides a convenient alternative solution for specific scenarios.

Future development of NumCraft: Short-term goals include supporting more machining types (turning, grinding, etc.), enhancing safety verification (integrating physical simulation), and optimizing the web interface; medium-to-long-term vision includes digital twin integration, adaptive machining, cross-platform support for industrial robot programming, and cloud-based SaaS services. The project is open-source, and community contributions are welcome (expanding knowledge bases, developing agent functions, improving NLP capabilities, etc.). We look forward to feedback from manufacturing users for continuous improvement.