Wealth-tech buyers do not ask “can we add a chatbot?” They ask whether generative AI can sit on portfolio and CRM context without creating a second warehouse of client ledgers. That is the data chasm: useful models want context; regulated operators cannot casually duplicate PII into a vendor cloud.
Open Portfolio’s answer is a BYOC inference boundary. Broker exports and approved aggregates are normalized at the edge; the enterprise keeps IdP and storage; inference receives a bounded, request-scoped context — not an open-ended retention of raw financial DNA. Pocket Portfolio is the live harness proving adapters under CSV chaos, not Open’s default enterprise store.
CTO and CISO diligence should separate three questions: (1) what is ingested and where, (2) what crosses the inference hop, and (3) what is retained after the response. Architecture pages and design-partner briefs exist so those answers are inspectable — not promised as certification theatre.
Use this pillar when evaluating generative use cases for portfolio commentary, exception triage, or operator assistants. Pair it with DORA / EU AI Act framing and the edge-vs-warehouse comparison before booking a diligence call.