api.xlsx-for-ai

Spreadsheets come back wrong more often than they come back broken. Most tooling reads a workbook as a grid of values, which works until the file has formulas — and then two things go wrong quietly. Values come back as whatever was last cached, which may be stale or missing entirely. And a round trip through a re-render drops everything that wasn't a cell value: charts, pivots, conditional formats, data validations, named ranges, styles.

Neither failure announces itself. The file opens, the numbers look plausible, and the analysis runs against them.

xlsx-for-ai recalculates in its own engine across 382 Excel functions, including dynamic arrays and XLOOKUP/FILTER, returning freshly computed values rather than cached ones. Edits are a structural splice into the existing workbook, so anything you didn't touch passes through untouched — and every write names the engine that produced the bytes, so you always know which path served you.

56 tools cover the round trip: read, inspect, recalculate, write, diff, clean, redact, and repair broken external references. It also opens formats that can't be uploaded at all — .xls, .xlsb, .ods, .numbers, Lotus, Quattro Pro — which is how an agent gets a legacy file in.

Workbooks up to 100MB and CSVs up to 200MB are handled server-side, past the point where loading a file directly stops being practical. Parsing is hardened against XXE injection, zip bombs, and entity expansion. Formula evaluation is sandboxed with network functions disabled, and recalculation runs entirely in our own path — no third-party formula engine sees your data.

Free accounts include a monthly file allowance; higher volume requires a paid plan.