Document Indexing 🟡 BETA

Documents in .gbkb folders are indexed automatically. No manual configuration required.

Automatic Triggers

Indexing occurs when:

• Files added to .gbkb folders
• Files modified or updated
• USE KB called for a collection
• USE WEBSITE registers URLs for crawling

Processing Pipeline

Document → Extract Text → Embed → Store in Qdrant

No chunking stage (September 2026). A document becomes a single record. There is no splitting into segments and no overlap, so precision degrades on long documents. This is a known limitation, not a setting.

Supported File Types

Website Indexing

Schedule regular crawls for web content:

SET SCHEDULE "0 2 * * *"  ' Daily at 2 AM
USE WEBSITE "https://docs.example.com"

Schedule Examples

Configuration

The embedding endpoint is part of the bot’s configuration. Two models are supported:

Input is truncated to 600 tokens before embedding with the local model. If no embedding endpoint is configured, retrieval degrades to keyword matching, and if embedding generation fails at query time a non-semantic hash vector is substituted — see Retrieval and RAG.

Using Indexed Content

USE KB "documentation"
' All documents now searchable
' LLM uses this knowledge automatically

Troubleshooting

See Also

• Knowledge Base System - Architecture overview
• Retrieval and RAG - How retrieval works
• Vector Collections - Collection management