Search
Subject search for AI systems
Search gives an AI system a current corpus on a defined subject, built from sources the organisation approves. It is not a search of the open web, and it is not another chatbot.
Each subject keeps its own index: the sites, feeds and pages that belong to that subject, refreshed on a schedule and retrieved when a model is allowed to look.
What you get
Search stays inside an approved subject
A slice such as cybersecurity, regulation or European AI infrastructure is configured by its sources and its terms. The engine does not try to index the general web.
No commercial search API in the path
Discovery comes from the sources you name: websites, RSS and Atom feeds, sitemaps, and URLs a person or another system submits. The corpus is yours to operate.
Models retrieve documents, they do not own the index
An allowed AI system can search and fetch a document. Assessment and answers stay with the model behind PEP and Router. Search does not need a model of its own.
The corpus stays current
A scheduled crawl refreshes the index and then stops. New items from feeds and sitemaps, and URLs submitted as seeds, enter the next run.
Crawling stays bounded
Each run has limits on pages, depth and time. The crawler respects robots.txt, and it rejects private networks, loopback and cloud metadata addresses.
Operators can see which sources earn their place
Sources can be added, paused or removed. Crawl health and source usefulness are visible, so an unused feed does not stay in the subject forever.
This page describes Search. It is not available to order.
The other products
- PEP — The boundary in front of the model. Corporate policy is the floor. A person can make that boundary tighter. They cannot make it looser unless the organisation has said so.
- Router — Policy decides which routes are allowed. Router decides which of those routes is the right one: sufficient capability, at the lowest expected cost.