Knowledge bases are an important part of artificial intelligence (AI) systems, especially for those agents or chatbots (Bots) based on large language models (LLM). Knowledge bases exist to make up for LLM's shortcomings in domain-specific knowledge and to solve the problem of "hallucinations" (i.e., fabricating facts) that may occur when they generate information.

The working principle of the knowledge base is that when an LLM-driven Bot needs to answer a user's question, it will not only rely on its pre-trained model, but also search for relevant information in its knowledge base. This knowledge base contains a wealth of proven facts and information

The advantage of a knowledge base is that it allows bots to provide more accurate and relevant information. When a user enters a question or request, the Bot can search the data in the knowledge base to find the most relevant information, then refine and summarize it and provide it to the user as an answer.

Through such a knowledge base, AI systems can better understand and process knowledge in specific fields, thereby providing higher quality services. This not only improves the user experience, but also enhances the credibility and practicality of the AI system.

Create a knowledge base

Embedding Model： The data vectorization processing model is used to vectorize data and cannot be changed after creation. Different models will have different results after vectorizing the data.

Q&A Split Model： It is used to generate question and answer pair data when AQ split is selected for importing data, but this model will not be used when other import methods are selected.

Upload documents

• Manual entry Manually enter question and answer pairs, which is the most accurate data
• Direct segmenting Select a text file and process it directly into segments
• Q&A Split Select a text file and let AI automatically generate question and answer pairs
• CSV Import batch import question and answer pairs

Paragraph Length

The length of each paragraph slice can be adjusted according to the actual situation

Customizable delimiter

Allows you to chunk based on custom separators. Usually used for processed data, using specific delimiters to accurately chunk it.

QA split guide words

In the QA split function, the guide word is a key parameter, which directly affects the quantity and quality of question and answer pairs generated by the LLM model. Therefore, please choose language that is relevant to the content as much as possible, and you can customize the lead words.

Knowledge base permission management

Corporate or public repositories permit access and use to authorized users. However, when the usage permission of the knowledge base is set to private, this function setting is invalid, and only users with permission can use or participate in maintaining the documents of the knowledge base.

Access permissions

There are two types of permissions to use the knowledge base:

• Public: Members within the company are authorized to use it.
• Private: only available to you

Limit on the number of knowledge bases

You can only create up to 6 personal knowledge bases, while the number of company knowledge bases depends on the number of packages purchased.