usecase
Market Research
Research agents that show their working
Market research produced by agents has a credibility problem: the conclusions arrive fast and the basis for them evaporates. CueCrux keeps the basis. Every finding an agent surfaces about competitors, market conditions, or trends carries a receipt to its source, so a landscape report circulates with its evidence attached and an executive can push on any claim and reach the material behind it.
The economics are the other half. Broad market sweeps burn retrieval tokens, and unbudgeted retrieval is how research programmes quietly become cost centres. Every CueCrux retrieval call takes a hard token budget and the daemon trims to fit. On the open ScoreCrux benchmark (Delta corpus, 2M+ tokens; methodology at github.com/CueCrux/ResearchCrux) this discipline returned far fewer tokens than naive top-K retrieval at near-identical answer quality.
Because runs are receipted and dated, refreshing the research is a rerun and a diff: what changed since last quarter is a comparison over signed records rather than a new project. Strategy teams get intelligence that is current, costed, and checkable.