Skip to main content

Knowledge Base

Integrating knowledge bases into workflows is recommended for scenarios such as: using outputs from previous plugins/LLMs as inputs for knowledge base retrieval; outputting knowledge base retrieval results to subsequent plugin/LLM nodes for processing (knowledge base node outputs are strings, generally requiring an LLM node after the knowledge base node to process retrieval results and generate responses, similar to knowledge base application Q&A).

1. Overview​

  • Defaults to using the previous node's output as this node's input (i.e., the question for knowledge base retrieval), can also select outputs from other nodes as input
  • Supports selecting created knowledge base(s) for retrieval
  • Supports setting knowledge base retrieval count and similarity threshold

2. Configuration​

  1. Node Input: Defaults to using the previous node's output as this node's input, can be manually changed to outputs from other previous nodes as input. Note: This node must be directly or indirectly connected to the start node to select previous node output parameters.
  2. Knowledge Base Selection: Select one or multiple knowledge bases. During execution, similarity retrieval will be performed in selected knowledge bases based on input content.
  3. Retrieval Method: Supports Semantic Retrieval or Enhanced Retrieval. Semantic retrieval uses text semantic vector similarity for retrieval, while enhanced retrieval combines semantic retrieval with full-text keyword retrieval. See Knowledge Base
  4. Similarity Threshold: Only knowledge base entries with similarity above this threshold will be retrieved. You can test in "Knowledge Base - Search Test", for example:
  • High similarity (e.g., 0.8+) will only retrieve knowledge highly relevant to questions, providing more accuracy but may miss results
  • Low similarity (e.g., 0.7-) may retrieve less relevant content but provides broader retrieval scope
  1. Semantic Retrieval Count: Maximum number of passages retrieved from knowledge bases through semantic retrieval per Q&A session, defaults to 3. Note this value shouldn't be set too high as it may exceed model context limits.
  2. Full-text Retrieval Count: Maximum number of passages retrieved from knowledge bases through full-text keyword retrieval per Q&A session, defaults to 1 when enhanced retrieval is enabled. The sum of full-text and semantic retrieval counts cannot exceed 10, with semantic retrieval count prioritized. If no keywords are matched during actual use, full-text retrieval results may be 0.

Note

Knowledge base nodes generally need to be followed by an LLM node to select, organize, and package knowledge base retrieval results before output, as shown: