Fancy UI KitFancy UI Kitv0.5
DocsLearnPackagesTUIFlowUse CasesInspirationShowcaseLeaderboard
Sign in
All use cases
Agents in your app

Have an agent build a spreadsheet the user can check

Model-generated workbooks, reviewed in the browser, exported as real xlsx.

The problem

Asking a model for "a budget spreadsheet" gets you a wall of CSV in a chat window. The user cannot check the formulas, cannot fix the one wrong cell, and cannot hand the result to a colleague who expects a real file.

You need the model's output to land somewhere a person can inspect and correct before it becomes a deliverable — and then to leave as a genuine xlsx, not a comma-separated approximation of one.

What you are building

Live surfaces, not screenshots — every one composed from the same components you would install.

Live workbook

The real SheetWorkbook with working SUM formulas — the surface an agent fills, not a picture of one.

The code

The agent writes cells, not a filets
// The sheet bridge exposes the workbook as controlled state, so an agent edits
// the SAME surface the human is looking at -- formulas recalculate live, and
// nothing has to be re-opened to see the result.
registerSheetsBridge(server, {
  adapter: {
    getWorkbook: () => workbook,
    setWorkbook,
  },
});
Export the finished thing server-sidephp
// holy-sheet writes real xlsx from the same data -- no headless browser, no
// spreadsheet application on the server. The Node twin (holy-sheet-js) writes
// the identical file if your backend is not PHP.
HolySheet::fromArray($rows)->write(storage_path('app/q1-sales.xlsx'));

How to solve it

  1. Put a real workbook on the page

    Render the sheet surface with the workbook in controlled state, so both the human and the agent write to the same object.

    Run thisbash
    npm install @particle-academy/fancy-sheets

  2. Bridge it

    Register the sheets bridge to expose the sheet_* tools. The agent writes cells and ranges through them, which is a normal state update — the human keeps editing the whole time.

  3. Let the human correct it

    This is the step that makes the pattern worth it. The agent produces a draft; the user fixes the two cells it got wrong. Neither has to redo the other's work.

  4. Write a real file server-side

    Export through the xlsx writer that matches your runtime — they are ports of each other, so the same workbook produces the same file whichever side runs it.

    Run thisbash
    composer require particle-academy/holy-sheet
# or, on Node:
npm install @particle-academy/holy-sheet