Upload the .docx or .xlsx your team already uses — with {{tokens}} where the content goes. Describe what you need in chat once, and get back a real Word document or Excel workbook: native tables, real numbers, your formatting, your branding.
Each generator shares the veilSense block system, adapted to what Word and Excel actually are.
Light markdown — bold, italic, and multi-paragraph or bulleted output when the block stands on its own paragraph — rendered as real Word formatting.
The template paragraph is cloned once per extracted item, however many the AI generates.
Admin authors one template data row; the LLM fills up to the configured max rows and columns.
The LLM chooses per block between a rendered Mermaid/PlantUML diagram and an AI-generated image, with an optional caption.
Fixed catalog assets for logos and letterheads; {{token}} strings (including date patterns) resolved without ever calling the LLM.
Optionally flagged numeric so downstream formulas — SUM, AVERAGE, whatever your template already has — total it correctly.
The template's data row is cloned to fit the LLM's output. Designated numeric columns are written as real numbers, not text.
Diagrams or generated images dropped directly into a sheet.
Same catalog-asset and token-resolution behavior as Word — no LLM call for boilerplate.
Every sheet gets its own purpose description shown to the extraction LLM, so a workbook with five tabs fills correctly in one pass.
A common failure mode for AI-generated spreadsheets is text where a number should be — and every formula downstream breaks silently. veilSense writes numeric-flagged cells as real numbers, so your existing SUM and AVERAGE formulas total AI-filled data correctly.
The Word and Excel Generators aren't a rich-text editor with an AI sidebar. You describe the report you need in a chat message, and the assistant fills the whole template in one pass — every block, every sheet, in one generation.
Set the template up once. Generate as many reports as you need afterward.
Your real template, with {{tokens}} marking where content goes — no separate shape-mapping step required.
Set each token's block kind, and for tables, the max rows and columns the LLM can fill.
One brief fills the whole document. Download it or save the result straight to Brain.
Map your template's tokens once. Let AI fill it — with real tables and real numbers — every time after.