Why Chat-with-Docs Breaks in Real Companies: An Engineering Look at Onyx

Chat-with-docs demos often work on static files but break in real companies where knowledge is scattered, permissioned, stale, deleted, or action-oriented. This article uses Onyx as an engineering case study to explain why trustworthy enterprise AI needs connectors, permission-aware retrieval, freshness pipelines, inspectable search, governed actions, and clear deployment boundaries.