Create Knowledge Base

The two most important choices when creating are the name and the embedding model. Content can be added later, but the naming boundary and retrieval method will affect future maintenance.

For the first try, you can set the [Embedding Model] to [Do not use]. The knowledge base will still use BM25 keyword retrieval, so just get the import and recall flow working first.

Make two decisions before creating

The name should describe the scope of the content

Prefer using “object + purpose,” such as [Employee Travel Policy], [Product After-Sales Manual], or [Market Research Materials]. Avoid names like [Materials] or [Test], which make it impossible to tell the content scope later.

Choose the embedding method

Choose	Suitable scenarios	Retrieval method	Prerequisites
Do not use	First try, keywords are clear	BM25 keyword retrieval	None
Cloud embedding model	The user's wording differs greatly from the original material	BM25 + vector hybrid retrieval	The corresponding model service can be called normally
local embedding model	Want vectorization to be done locally	BM25 + local vector retrieval	First complete the download in [Local Models]

Creation steps

1

1. Open the creation window

Open the [Knowledge Base] in the left navigation, then click the Add button above the knowledge base list.

2

2. Enter a name

Enter a name that indicates the scope, such as [Employee Travel Policy].

3

3. Select an embedding model

Choose an available cloud or local embedding model; if semantic retrieval is not needed for now, select [Do not use].

The name determines the scope of the content; the embedding model determines whether vector retrieval is included.

4

4. Click Create

After confirming the name and model, click [Create]. Once creation is complete, you will enter an empty knowledge base.

5

5. Add the first batch of materials

Click the add materials button, import one or two files or notes with clear answers, and wait for processing to complete.

Use a local embedding model

Open [Settings] → [Local Models], and download an available model in the [Embedding Model] section. The models shown in the interface and their download sizes may vary depending on the installation environment; use the current list as the reference.

After the download is complete, return to the knowledge base creation or settings page and select this model.

Local embedding only means vectorization is completed on this machine. Whether document parsing, reranking, and chat use the cloud still depends on the services and models you choose.

Change the model after materials already exist

When enabling an embedding model for a knowledge base that previously used only BM25, you can build a vector index. When changing the embedding model for a knowledge base that already has vectors, the interface will enter the [Rebuild Knowledge Base] process.

Before starting the rebuild, make sure the new model can be called normally. After rebuilding, run the recall test again; do not change the model and modify the chunks in the same round, or you won’t be able to tell where the result changes came from.

Configuration notes

Configuration item	Product default	Suggested starting point	Effect	Applicable scenario	Notes
Name	Empty	Object + purpose	Differentiate the scope of materials	All knowledge bases	Materials with different permissions or lifecycles should be separated
Embedding model	Do not use	Do not use it for the first try	Determines whether vector retrieval is added	Colloquial phrasing and many synonymous expressions	Billing and data handling for cloud models depend on the service provider
local embedding model	Not downloaded	Download only when local processing is needed	Complete vectorization on this machine	Offline or higher privacy requirements	Parsing, reranking, and chat models still need to be checked separately

Expected result

• The new knowledge base appears in the list, and its name can be distinguished from other knowledge bases.
• You know whether the current retrieval is keyword retrieval or hybrid retrieval.
• The selected cloud model can be called, or the local model has finished downloading.

User case

Xiaolin created an [Employee Travel Policy] knowledge base for the first time. He first chose [Do not use] for the embedding model, imported three policy documents, and completed the recall test. After the keyword queries became stable, he configured an embedding model and compared the results of colloquial queries using the same questions.

The completion standard is: after upgrading the retrieval method, the original fixed questions do not degrade, and colloquial queries can more consistently find the same policy.

Common questions

What should I do if the Create button is unavailable?

Check whether the name is empty and whether the selected model is still available. If the model service is not configured, you can first switch to [Do not use] to complete creation.

Will not using an embedding model mean I can’t search anything at all?

No. The knowledge base will still use BM25 keyword retrieval; the closer the question wording is to the materials, the more stable the results usually are.

Do I need to create a separate knowledge base for each topic?

Use “should these be retrieved together when in use?” as the criterion. Materials with completely different permissions, lifecycles, or topics are more suitable to keep separate.

Continue reading

Add and organize materials

Import content and handle same-name conflicts.

Check materials and recall

Validate results with real questions.

Model and retrieval settings

Learn about embeddings, reranking, and rebuilding.