The questions
that come up in every call.

Bloomberg and FactSet are read-only data terminals — you do the modelling somewhere else, and the framework you use to do it lives in your head. Excel plus a copilot is a chat layer over a workflow that doesn't survive turnover; when the analyst leaves, the model goes with them.

Nagini encodes your framework once — how you size revenue, how you bunch opex, what disqualifies a name — and runs it forever after. The agent operates in your shape, not a generic one. Re-runs the model on every filing, surfaces what changed, hands back a workbook a PM can read.

Signed DPA on day one. Your data never trains any model — not ours, not a vendor's. Single-tenant available at the institutional tier. The trust block on the home page is the short version; the longer version is on the call.

One prompt ships your full cell graph as a styled .xlsx — formulas where they make sense, headers and formats already wired. The workbook reads as if a person built it. The walking-away case is the one we optimize for, not the lock-in case.

No. The agent reads cells, writes cells, runs DCFs, and exports workbooks. It does not touch a brokerage. Trade execution is regulatory-gated and not on the near roadmap.

First token in under 2s on most prompts. A full reply with a DCF re-run takes roughly 15-30s — the worker round-trip dominates, not the LLM. Concrete numbers from the live walkthrough; they get tighter as caching lands.

Not yet, on purpose. Export is shipped; import is parked. A half-baked importer rebuilds the workflow Nagini is meant to replace. The importer ships when the workspace can hold its own — not as a wedge, not as a hedge. In the meantime: hand a thesis to the workspace and let the seeded cells start there.