This paper introduces a method for developing an interactive CAD intelligent assistant based on the local large model deployed by Ollama ^[1] and the AnyCAD three-dimensional graphics platform ^[2] . The method realizes the function of assisting the establishment of two-dimensional and three-dimensional geometric models based on the large model. The main contents include the deployment of the Ollama local large model, the interaction between AnyCAD and Ollama based on WebSocket, and the model interaction based on the Model Context Protocol (MCP).

1 Introduction

1.1 Ollama local large model deployment

Ollama is a high-performance, large-scale machine learning model framework that supports deployment in local environments.

In order to deploy Ollama in a local environment, you first need to make sure that your server or workstation meets the hardware requirements for running Ollama. This article uses i9-13900K + 64GB memory + 4080S16G, based on the qwen2.5 14b large model.

Ollama installation and deployment of qwen2.5 is briefly described.

1.2 Model Context Protocol MCP

The Model Context Protocol (MCP) defines a standard format for passing information between applications and machine learning models. It enables data from different sources to be understood in a unified way, thus simplifying the model application process.

1.3 AnyCAD Graphics Platform

AnyCAD graphics platform provides APIs such as C++, .NET, Python, and Java, which can facilitate the development of CAD/CAE/CAM applications. Its concise API ^[3] is very suitable for calling large AI models. A wealth of examples ^[4] are provided on gitee , which can help you get started quickly. AnyCAD .NET SDK ^[5] has been downloaded more than 300,000 times on nuget.

2 Overall Architecture

The overall system architecture is shown in the figure below:

in:

• • AI smart assistant

Based on the AnyCAD graphics platform API, the client plug-in of the intelligent assistant is implemented, which mainly communicates with the AI ​​Server through the WebSocket protocol.

• • AI modeling service plug-in

Based on AnyCAD graphics platform API, the plug-in of the intelligent assistant is implemented to implement the MCP protocol.

• • AI Server

Interact with LLM through the API provided by Ollama, receive AI smart assistant requests through WebSocket and return AI model responses.

3 Effect display

•  AI assistant 2D drawing

4 Summary

Through simple verification, the process of developing an AI smart assistant has been completed. However, more work and in-depth optimization are needed to achieve a smarter CAD smart assistant for engineering applications.