Managing the Hi, I'm Kai knowledge base
This page explains how crawled pages, inline context, source controls, and generated structure combine into the knowledge base Hi, I'm Kai uses for answers.
What is the knowledge base?
The knowledge base is the body of content Hi, I'm Kai draws from when answering visitor questions. It is automatically populated by crawled content, but you can refine it manually.
Each included source becomes part of the retrieval layer. Better source quality usually means better answer quality.
When should I use inline context?
Inline context is a short text summary that the AI always has access to without needing vector search. Use inline context for high-priority information such as business hours, pricing summaries, service boundaries, or brand positioning.
Keep inline context concise. The token counter in the editor shows real-time usage against the maximum allowed so response latency stays low.
How do sources and pages work?
Each crawled page becomes a source in the knowledge base. You can include or exclude individual pages, prioritize key URLs, and manually add content.
Use source controls to remove outdated pages, low-quality pages, or pages that should not influence answers. Keep public, authoritative pages included.
How does structure generation improve retrieval?
Use Generate structure to let Gemini analyze your crawled pages and produce a semantic summary that improves retrieval accuracy.
Generated structure helps the system understand important topics, relationships, and page roles. It is especially useful when a website has many pages that cover related concepts.