NexusAI AI Advisor is a read-only analysis feature. The current request path sends the user’s question plus selected source-backed reporting evidence through the Vercel AI Gateway to the configured model provider.
The question, dataset label/key, source filename, bounded reporting observations, quality counts, period comparisons, and allowed evidence references needed to answer the question. The current Advisor path does not upload the original spreadsheet file to the model.
Only after a signed-in user with Reporting + AI access asks the Advisor a question using a source-backed evidence packet.
Nothing through the current Advisor endpoint. The evidence packet must be marked read-only, and the endpoint returns analysis, facts, suggestions, assumptions, and limitations. It has no write action in that request path.
Evidence is bounded and sanitized. Returned evidence references are accepted only when they already exist in the supplied packet. If evidence is insufficient, the Advisor is instructed to say so rather than invent facts.
Provider-side retention and training treatment can depend on the gateway/provider account and policy in effect. NexusAI does not promise zero retention or no training unless that production provider configuration is independently verified.
See Security for the broader control summary.