Aaron Linsky and Sam Green said the system contains more than 100 million documents from 135 external sources alongside internal material. Documents are enriched with the markets, concepts, and time periods they cover.
The team described Vault as a shared research layer for human investors and agents, with permissions and time-aware retrieval. In a demonstration about AI infrastructure investment and the Korean won, a Vault-connected agent restricted its search to October 1–8, found 52 candidate documents, and used seven. The session compared this with a public-web chat client that found 15 plausible sources and used one or two.
Bridgewater said it manually labeled roughly 200 questions to evaluate search, and that precision at ten rose from about 60% to roughly 90–95% after subsequent improvements, including agentic search. It also described saved searches, collections, and research findings being returned to Vault for later work. The source presents a company demonstration, not a general benchmark or evidence of investment performance. It does not establish external availability, pricing, or customer access to Vault.