Blog
Product updates, engineering insights, and company stories from the team building the future of web research for AI.
BFS crawling wastes 95% of your page budget on navigation. We replaced it with a single sentence of intent — and found 87 targets where BFS found 2.
How we built a two-pass signal scoring system that suppresses 75% of notifications, catches subtle semantic changes, and self-heals when monitors drift — with architecture details, scoring mechanics, and production data from 9,000+ patrol runs.
How we turn a plain-English monitoring intent into an executable workflow graph, run it on a schedule, detect meaningful change via URL set algebra and content hashing, and repair monitors that drift, with architecture details from 12,000+ production monitor runs.
Knowledge engines give agents recall over static data. Persistent monitoring gives them awareness of live change. This post explains what makes perception architecturally different from retrieval, what we tried building it, and what broke along the way.