Build a Knowledge Base with Dify in 5 Minutes – Large Model Learning Part 2

Installation

The tutorial begins by installing Docker Desktop from the official Docker website, then cloning the Dify repository at https://github.com/langgenius/Dify. In the dify/docker/ directory, run the following commands in PowerShell:

cp .env.example .env
docker compose up -d  # add --verbose for detailed logs if needed

If network issues cause Docker pull failures, the guide suggests configuring domestic mirror sources by editing the Docker daemon JSON with entries such as "https://docker.m.daocloud.io", "https://hub-mirror.c.163.com", "https://mirror.baidubce.com", and "https://docker.nju.edu.cn". After a successful start, open http://localhost/install in a browser, where the UI shows four menus: Exploration, Studio, Knowledge Base, and Tools.

Adding a Local Model

To use a local model, the tutorial downloads a vector model with Ollama:

# Download vector model
ollama pull bge-large
ollama list

The user then selects the model provider ollama in Dify’s settings and adds the model.

Creating a Knowledge Base

In Dify, click Knowledge Base → Create Knowledge Base . The default segmentation settings are used, but the guide explains each option:

Segment Identifier : default \n, splits text by paragraph.

Maximum Segment Length : default 500 Tokens, with an upper limit of 4000 Tokens.

Overlap Length : recommended 10‑25% of the segment length to improve recall.

Pre‑processing Rules :

Replace consecutive spaces, newlines, and tabs.

Delete all URLs and email addresses.

After configuring, Dify provides a real‑time preview of the segmented chunks, allowing users to adjust the strategy on the left‑hand preview pane.

Indexing and Retrieval Settings

The tutorial advises using the recommended indexing and retrieval options without further modification.

Testing the Knowledge Base

To verify the setup, create a blank application in the Studio, choose the Chat Assistant template, and add a new text entry such as “张江吴彦祖是技术豆”. Select the previously created knowledge base as the source. Sending a query shows that the assistant can retrieve the inserted content, confirming the knowledge base works.

Next Steps

The author notes that the next episode will cover building workflows with Dify.