The painful version of this is one shared index, where one company can surface another company’s files. We keep each tenant’s documents apart, return the source with the answer, and leave you with a small set of questions you can re-run, plus a view of what it costs.
This sits inside the AI features we add to a product that already ships. The pattern for tenant boundaries is written up in the architecture note.
Only if you want them to see each other’s files. We do not ship that. Each tenant’s documents stay apart, and we check it before launch.
Answers that cite a source, a small set of questions you can re-run, and a view of what the feature costs.
