How to deploy a large enterprise-level model knowledge base based on a MacBook

Written by

Iris Vance

Updated on:June-13th-2025

— 1 —

Why build a knowledge base on your MacBook?

The most important thing is that due to security requirements, the documents we have cannot be uploaded to the cloud service at will, so we cannot actually verify the actual effectiveness of the knowledge base. In addition, for IT students, building a complete solution by themselves, being able to flexibly adjust and connect various models, and evaluating the different performances of various models are also driven by their instinctive exploration of technology.

The configuration of the MacBook used is as follows. After the large model is quantized (for example: int8), it can run smoothly.

— 2 —

Knowledge base architecture design

The architecture design based on a MacBook is shown in the figure below. In this architecture design, we adopted the domestic open source large model Qwen3-14B, which is upstream and friendly to both enterprises and academia.

The embedding search RAG solution based on the m3e-base model was adopted. Based on these two models, an API interface protocol compatible with ChatGPT was encapsulated; by introducing the One API interface management & distribution system, a unified LLM interface channel management platform specification was formed, and the encapsulated interface protocol was registered; FastGPT, an open source large model knowledge base platform management system on par with dify.ai, was built to realize a complete knowledge base application process integrating private knowledge data source preprocessing, embedding retrieval, and large model dialogue.

Although it is small, it has all the necessary functions, and finally formed a solution that meets commercial standards and can run on MacBook. Although there is still a certain gap between the intelligence level and actual needs, at least we can deploy and run it at the lowest cost without purchasing additional graphics cards or cloud services, and can import actual business data (such as design documents, business process documents, project management documents, technical sharing documents, etc.) for practical verification. It is worth every engineer to try it .

— 3 —

Knowledge base deployment design

The deployment solution based on MacBook is divided into four main links and 14 specific steps. As long as you practice step by step, every IT student can have his own private large model knowledge base system on his MacBook. The steps are as follows:

Some steps can be completed simply by deploying the Docker image with one click, but in line with the deployment idea of paying attention to details, a purely manual method was adopted.