Quickly Build a Personal Knowledge Base with DeepSeek – Step‑by‑Step Guide

Background

Why do you need a personal knowledge base? In daily work and study we accumulate many documents, code snippets, and notes. Without a unified storage and retrieval system, these resources become disorganized and time‑consuming to locate. This guide shows how to quickly build a personal knowledge base with DeepSeek, enabling local document upload, vectorization, and intelligent assistant replies.

Preface

The previous article covered local deployment of DeepSeek, which can be sluggish on a typical PC (16 GB RAM, 10 GB GPU) when running a 14B model, so local deployment of a knowledge base is not recommended. Instead, we will use the web version of DeepSeek.

Demo Effect

(Illustrative screenshots omitted for brevity.)

Building the Knowledge Base

1. Install Cherry Studio

Download from https://cherry-ai.com/download.html . If installation errors occur, the required packages are available on the provided Baidu Cloud link.

2. Knowledge‑Base Tutorial

The official documentation provides a macOS tutorial; the Windows tutorial is reproduced here for convenience.

Implementation Steps

1. Add a Chat Model

In the lower‑left settings, click “Add” and select the model name (e.g., deepseek-ai/DeepSeek‑R1).

2. Log In / Register on SiliconFlow

Visit https://cloud.siliconflow.cn/ . New users receive 20 million tokens for free.

3. Generate an API Key

After logging in, generate and copy the API key.

4. Configure the API Key in Cherry Studio

Return to Cherry Studio, paste the API key, and select the corresponding model.

5. Add an Embedding Model

Embedding models convert text into numeric vectors, which are the core of a knowledge base. The free BAAI/bge‑m3 model is used here and does not require an API key.

6. Create a New Knowledge Base

In Cherry Studio’s left toolbar, click the knowledge‑base icon.

Click “Add”, name the knowledge base, and start creation.

7. Add Files and Vectorize

8. Add an Assistant and Bind the Model

Create a knowledge‑base assistant.

Configure it to use the previously added chat and embedding models.

9. Query the Knowledge Base in Conversation

10. Verify the Result

The assistant returns answers that match the content of the uploaded documents.

Conclusion

If personal data does not raise privacy concerns, the web version is recommended because local deployment may be sluggish on modest hardware.

Building your own knowledge base greatly improves retrieval efficiency.

Feel free to ask questions and discuss further.