MOGICK: A Codex‑Style AI Agent that Automates Full‑Cycle Equipment Design

Background

After the emergence of CLI agents such as Claude Code and Codex, many have tried to use them to create CAD models, but their origins in AI programming limit their grasp of industrial knowledge, making them unsuitable for professional product design.

Introducing MOGICK

MOGICK is Tongyuan SoftControl's newly released high‑level AI agent aimed at equipment R&D. It interacts via natural language, similar to Codex, and sequentially completes:

Requirement understanding and system architecture design : defines required subsystems/components, performance metrics, and relationships.

Subsystem/component architecture and geometric prototype : creates parametric geometry models and assembly relationships.

System‑level simulation verification : runs simulations on the whole equipment model to find optimal subsystem parameters.

Case Study: Mars Sampling Mission

Using the official example, MOGICK first performed mission requirement analysis, then designed the mission architecture, simulated the flight trajectory, built the lander’s geometric prototype and architecture, and finally carried out multi‑discipline coupling and system‑level joint simulation.

System Architecture

The user layer is the MOGICK APP, a Codex‑like integrated interface for interaction and result viewing. Beneath it lies an Agent engine that executes various tasks, supported by a cloud platform for continuous backend execution and a plugin manager for integrating additional software.

Tool Integration

When creating system models, 3D geometry prototypes, or simulation models, MOGICK invokes the corresponding Tongyuan tools:

System modeling and viewing: SysBuilder

Geometric prototype modeling and viewing: SysCAD

Control algorithm modeling and simulation: SysLab

System joint simulation: SysExplorer

Agent Engine Mechanisms

The engine achieves an automated closed‑loop from task understanding to artifact generation and verification through multiple mechanisms: concurrent multi‑tool execution, session management, memory system, code and model search, sandbox execution, Git worktree management, and a goal‑oriented mode.

Quality Evaluation System

Output evidence includes intent, operation, semantic changes, CI/simulation acceptance conclusions, and confidence scores. A six‑dimensional assessment, A/B comparison, and attribution analysis form a comparable, traceable, and regressable quality baseline for evaluating results.

Limitations and Future Outlook

MOGICK focuses on system‑level design, determining parameters and performance requirements for subsystems/components, which serve as inputs for detailed geometric design. In theory, plugins could connect professional CAD/CAE tools for detailed design, but the complexity of detailed geometry suggests future specialized agents will be needed to receive high‑level goals from MOGICK and perform detailed design.

Currently, MOGICK itself does not possess deep domain knowledge; professional industrial design knowledge, standards, skills, and tools must be added as extension packages to enable AI to truly complete complex equipment design.

Intelligent Equipment R&D Paradigm

MOGICK exemplifies a future paradigm where AI‑driven equipment system engineering evolves from point‑wise assistance to an intelligent closed‑loop across five maturity levels:

L1 Question‑Answering (Chatbot) : AI assists engineers with knowledge queries and information retrieval.

L2 Assistance (Copilot) : AI enters modeling environments to suggest parameters, complete designs, and check errors.

L3 Task (Agent) : AI can execute specific tasks such as requirement analysis, architecture design, and model construction.

L4 Engineering (Agentic Engineering) : AI orchestrates the entire engineering workflow—requirements, design, simulation, testing, and validation—understanding relationships among models, parameters, interfaces, constraints, and context.

L5 Autonomous : AI achieves end‑to‑end system engineering, autonomously delivering architecture, models, simulation validation, and optimization from requirement input.

L4‑Level New Equipment System Engineering Framework

The L4 framework consists of four core elements:

Unified multi‑paradigm model representation : Treat design as modeling and models as code, using rigorous semantic specifications and formal methods to express diverse engineering models uniformly.

Industrial knowledge and large‑model foundation : Aggregate requirements, standards, models, cases, and experience into a searchable, verifiable, traceable knowledge base; employ RAG, continual training, and feedback loops to create a data‑closed‑loop that tightly couples with digital engineering toolchains.

Equipment engineering AI agents : Blend domain knowledge with engineering semantics to orchestrate modeling, simulation, optimization, and verification, preserving reusable evidence.

AI‑native industrial software suite : Provide AI‑native tools for requirement analysis, architecture design, scientific computing, system modeling and simulation, geometric design, and specialized simulation.

Current Status

MOGICK has been announced but is not yet publicly available; a launch is expected soon. Interested readers can consult the original announcement for further details.