Train Chat/AI on website rather than links
released
Taylor Jennings
Training Chat AI on website content currently requires adding URLs one by one. There is no way to point the system at a domain and have it ingest content automatically.
Admins should be able to provide a root domain and have Chat AI automatically scan and train on all relevant website content, rather than adding individual links manually.
Manual link-by-link training doesn't scale for large or frequently updated websites. Automated ingestion reduces setup time and ensures the knowledge base stays current as content changes.
Taylor Jennings
marked this post as
released
When adding a website resource, you now have three modes:
- Whole website – Enter a URL and choose a scan depth: Shallow (1 level), Medium (2 levels), or Deep (3 levels). Previously this was locked to 2 levels. Once the scan runs, results load progressively and are displayed in a tree – so you can drill down by path and selectively import whole sections (e.g. everything under /resources/customer-stories) or individual pages.
- Sitemap – Unlike Whole website, which stays scoped to the domain you enter, Sitemap reads your sitemap and picks up subdomains too (e.g. help.chilipiper.com, fire.chilipiper.com). One scan captures content across your entire web presence.
- Specific webpages – No change, comma-separated URLs as before.
See more: https://cleanshot.com/share/DG07HGnX
Taylor Jennings
Merged in a post:
Automatically train Chat AI on website content without adding links manually
Taylor Jennings
A more automated way to train the AI on website content, instead of adding links one by one
Taylor Jennings
Merged in a post:
Improved Knowledge Base Management for Chat AI
Taylor Jennings
A more efficient method for integrating support documentation into the chat AI, as the current requirement to link each article separately is cumbersome.