Chapter 03 – Knowledge Base System Overview

This chapter explains how botserver manages knowledge base collections, indexing, caching, semantic search, and conversation memory. The implementation uses vector databases for semantic search and intelligent memory management for context optimization.

Document	File	Description
README	README.md	High-level reference for the `.gbkb` package and its core commands (`USE KB`, `CLEAR KB`, `USE WEBSITE`).
KB and Tools	kb-and-tools.md	Integration patterns for knowledge bases and tool systems.
Vector Collections	vector-collections.md	Definition and management of vector collections, including creation, document addition, and usage in dialogs.
Document Indexing	indexing.md	Process of extracting, chunking, embedding, and storing document vectors in the VectorDB.
Semantic Search	semantic-search.md	How semantic search performs meaning-based retrieval using vector embeddings.
Episodic Memory	episodic-memory.md	Automatic conversation history management, context compaction, and intelligent summarization.
Semantic Caching	caching.md	Intelligent caching for LLM responses, including semantic similarity matching.

Key Configuration Parameters

Knowledge Base

Parameter	Default	Description
`embedding-url`	`http://localhost:8082`	Embedding service endpoint
`embedding-model`	`bge-small-en-v1.5`	Model for vector embeddings
`rag-hybrid-enabled`	`true`	Enable hybrid search
`rag-top-k`	`10`	Number of results to retrieve

Episodic Memory

Parameter	Default	Description
`episodic-memory-enabled`	`true`	Enable/disable episodic memory
`episodic-memory-threshold`	`4`	Exchanges before compaction
`episodic-memory-history`	`2`	Recent exchanges to keep
`episodic-memory-model`	`fast`	Model for summarization
`episodic-memory-max-episodes`	`100`	Max episodes per user
`episodic-memory-retention-days`	`365`	Days to keep episodes
`episodic-memory-auto-summarize`	`true`	Auto-summarize conversations

LLM Cache

Parameter	Default	Description
`llm-cache`	`false`	Enable/disable response caching
`llm-cache-ttl`	`3600`	Cache time-to-live in seconds
`llm-cache-semantic`	`true`	Use semantic similarity matching
`llm-cache-threshold`	`0.95`	Similarity threshold for cache hits

How to Use This Overview

Navigate: Click the file links to read the full documentation for each topic.
Reference: Use the parameter tables for quick configuration lookup.
Update: When the underlying implementation changes, edit the corresponding markdown files and keep this summary in sync.

3.2 KiB Raw Blame History Unescape Escape