| .. | ||
| caching.md | ||
| context-compaction.md | ||
| indexing.md | ||
| kb-and-tools.md | ||
| qdrant.md | ||
| README.md | ||
| semantic-cache.md | ||
| semantic-search.md | ||
| summary.md | ||
| vector-collections.md | ||
gbkb Reference
The knowledge‑base package provides three main commands:
- USE_KB – Loads and embeds files from the
.gbkb/collection-namefolder into the vector database, making them available for semantic search in the current session. Multiple KBs can be active simultaneously. - CLEAR_KB – Removes a knowledge base from the current session (files remain embedded in the vector database).
- ADD_WEBSITE – Crawl a website and add its pages to a collection.
Example:
' Add support docs KB - files from work/botname/botname.gbkb/support_docs/ are embedded
USE_KB "support_docs"
' Add multiple KBs to the same session
USE_KB "policies"
USE_KB "procedures"
' Remove a specific KB from session
CLEAR_KB "policies"
' Remove all KBs from session
CLEAR_KB
The vector database retrieves relevant chunks/excerpts from active KBs and injects them into LLM prompts automatically, providing context-aware responses.