The knowledge base is a key component for implementing RAG (Retrieval-Augmented Generation). It manages the import, processing, organization, and maintenance of knowledge. A knowledge base not only provides LLMs (Large Language Models) with dynamic, real-time context but also helps enterprises improve internal information retrieval efficiency, ensuring accuracy and timeliness of answers and generated results.
The knowledge retrieval process can be summarized in three steps:
1. Knowledge Import and Processing: Import files, Q&A, or database information into the knowledge base. The system then performs segmentation, indexing, and vectorization.
2. Keyword Search and Content Retrieval: When the LLM receives a user query, the knowledge base recalls the most relevant content chunks through keyword or semantic search and provides key context.
3. Context-Enhanced Generation: The LLM combines the recalled context with its own knowledge to generate answers, ensuring accuracy, timeliness, and business relevance.
Depending on the usage level, knowledge bases can be divided into Team-level and Application-level:
Team-level knowledge base: Under the same primary account, a knowledge base can be referenced by multiple applications, and one application can also reference multiple knowledge bases. You can create and manage them centrally in the Knowledge Base module, then reference them in specific applications.
Application-level knowledge base (Knowledge Management): Each application has one default knowledge base, which can only be used within that application and cannot be shared across applications. For details, see Knowledge Base Settings.
Creating a Knowledge Base
Click Knowledge Base>Create Knowledge Base in the left-hand menu, fill in the name and description.
Managing the Knowledge Base
In the Tencent Cloud Agent Development Platform, stored data within the knowledge base is referred to as knowledge. It is divided into three types:
Files: Structured or unstructured files such as PDF, DOCX, TXT, or web pages. Documents are segmented and indexed, with context slices stored in the knowledge base for retrieval.
Q&A: Knowledge in the form of question–answer pairs. The system automatically generates them from documents or manual input, enabling quick responses to common questions and improving coverage and efficiency.
Database: Structured data sources such as internal systems, tables, or API data. These can provide precise information for Q&A systems and LLMs, or serve as auxiliary retrieval conditions to enhance business-specific support.
Modify knowledge base name: Click the top-left of the knowledge base page to edit its name.
Note:
1. Document comparison tasks and tag management work only at the knowledge base level and cannot be shared across knowledge bases.
2. Knowledge bases support importing files, Q&A, and databases. Once knowledge is in the “Imported” state, it becomes effective.
Deleting a Knowledge Base
To delete a knowledge base, ensure it is not currently referenced by any application. If it is, remove the reference relationship first before deleting.
Referencing a Knowledge Base in Application
Click the app, go to App Management-Knowledge, and click Add Knowledge Base to establish a reference between the knowledge base and the application. After adjusting the reference relationship, you must republish the application. For details on adjusting knowledge management and segmentation settings, see Knowledge Base Settings.
