Web search

Web search is the one knowledge source you don’t supply. Everything else an assistant knows is material you uploaded; with web search it goes out and reads the page as it stands right now.

To switch it on for a particular assistant, see Web search under Abilities. This page covers what it can reach, what leaves Intric when it runs, and what it is replacing.

The whole web, or only the sites you allow

The allowed-sites list on the assistant decides the reach:

• Empty list — the assistant searches the open internet, the way you would with a search engine. Use this when you can’t predict where the answer lives.
• One or more URLs — the assistant searches inside those sites and nowhere else. Use this when you already know which source should be authoritative: a regulator, a standards body, your own public website.

The list is a boundary, not a preference. An assistant restricted to three sites cannot reach a fourth, however the question is phrased — so if a site must be out of reach, the list is what puts it there. Instructions written into the assistant only steer it.

What is sent when a search runs

When web search runs, the assistant’s model generates a search query, which is sent to Intric’s search sub-processor. Your original prompt, chat history, attached files and personal data aren’t sent to it — but the generated search query may reflect sensitive context from the conversation, so an appropriate security classification still matters. See below.

Web search replacing Crawl in old Knowledge

The Websites (Crawl) function in the older Knowledge area does something that looks similar, but works the other way around: a crawl reads a site in advance and keeps an indexed copy, which is why its contents go stale between runs. Web search reaches the live page at the moment of the question, so there is no copy to schedule and nothing to age.

Crawling is being retired, and web search is what replaces it. No date has been set for that changeover yet — existing crawled websites keep working in the meantime, and nothing you have already configured stops overnight. For the wider retirement timetable of the old Knowledge area, see Legacy knowledge.

If you are setting something up today, reach for web search restricted to the relevant sites rather than a new crawl. You get the same “answer from these pages” behaviour without a copy to keep fresh, and it is where the platform is heading.

Security classification and web search

Because the generated search query can reflect sensitive context, web search — like any tool — is subject to the security classification of the Space it’s used in. Your organization can block it entirely in Spaces classified above the level web search itself has been approved for.