universal agentic searcher

a deep research pipeline where an AI agent doesn't just retrieve results but actively iterates on its understanding — querying, synthesizing, identifying gaps, and querying again. the core insight is that good human research is a conversation with information: you search, find something, realize what you actually need to know, search differently, and gradually build a coherent picture. standard search is step-one-and-done. the agentic searcher loops: it generates an initial query, summarizes what it found, identifies what's missing or contradictory, generates follow-up queries, and repeats until it's confident it has a complete answer or until it surfaces its remaining uncertainty explicitly.

the adaptive feedback piece is what distinguishes this from just chaining search calls. the agent maintains a model of what it knows and what it doesn't — essentially a knowledge graph of the research question — and uses that to drive the next query. it also asks the user targeted clarifying questions when the research direction is genuinely ambiguous, rather than picking an interpretation silently. the output isn't a list of links but a synthesized answer with explicit sourcing, confidence levels, and a statement of what couldn't be verified. this is closer to how a good research assistant operates than any current search product.