Docs

Everything you need to set up, configure, and get value from the Hi, I'm Kai conversation layer.

Managing the Kai knowledge base

This page explains how raw source pages and editable inline context combine into the knowledge base used by the current beta.

What is the knowledge base?

Each site stores two related forms of content:

  • Raw source pages contain a URL, title, description, headings, body, and inclusion setting.
  • Inline context is prompt-ready text assembled from included pages and injected into chat prompts before on-demand retrieval.

The current beta does not expose a vector-embedding or chunk-management interface. Crawls and manually fetched pages update the raw list, while the editable inline context is the content you can inspect and save directly.

How is inline context built?

The builder starts with an Available pages index, then stuffs full page sections into a 75,000-token budget. It orders the homepage and common top-level business pages such as About, Pricing, Services, Products, Contact, FAQ, and Support first. Remaining included pages follow in URL order when they fit.

The Overview panel shows raw, included, and stuffed page counts plus an estimated token count. The estimate uses roughly four characters per token; the builder's hard budget is 300,000 characters.

Select Rebuild to regenerate the editor from the current included pages. Rebuilding does not persist the result until you select Save. You can also edit Inline context directly and save it without rebuilding.

How do I manage raw pages?

Select a page in Raw source pages to edit its URL, title, description, headings, or raw body content. The checkbox controls whether the page participates in inline context and retrieval. Exclude outdated, duplicated, or low-quality pages, then select Rebuild and Save.

To fetch a public URL outside the latest crawl:

  1. Enter the URL in Add page URL.
  2. Select Fetch & add.
  3. Review the fetched page and rebuilt context.

The same per-site page limits apply here: 50 Free, 500 Growth, and 5,000 Scale. Adding a new page at the limit is rejected. Refreshing an existing fetched URL is a manual sync and therefore requires Growth or Scale.

What does Generate with Gemini do?

Select Generate with Gemini to have the server rebuild and save inline context from the included raw pages. At least one page must be included. The result and normalized source list replace the current values and are synced to the live widget configuration.

This is a context-generation action, not a claim that the dashboard has built a separate vector index. Review the result before relying on it for visitor answers.

How does content sync work?

All plans display daily content checks. Growth and Scale add manual crawl and refresh controls. The Content sync panel shows Sync frequency, Last checked, Next check, and Knowledge updated.

Saving or generating context also invokes configuration sync. A success banner says the knowledge base was saved and synced; if CDN/KV publishing is unavailable, the dashboard reports that the database save succeeded and explains why live sync was skipped.