Is there an Excel Copilot alternative that keeps my file local?

What is Copilot in Excel built for?

Copilot is the AI assistant built into Excel, and its best argument is that it is already there. It writes and explains formulas, builds charts and PivotTables, applies conditional formatting, and answers questions about the numbers on the sheet, surfacing trends and outliers you did not ask about. Its analysis mode goes further than most people expect: it writes Python against your data and can leave the code in the workbook as a refreshable cell. And it edits in place — you are not copying results back from a chat window, you are watching your own spreadsheet change.

For the person whose working life is already inside a licensed Microsoft 365 tenant, that is a strong position. There is nothing new to adopt, nothing to install, and no new place your data has to live that it did not live already. If the question is "write me the formula that does this", nothing here will beat it.

Where does Copilot fall short for a quick, private question?

Start with where the file has to be. Copilot feels local because Excel is local, but Microsoft's own FAQ says Copilot in Excel needs the workbook saved to OneDrive, and the analysis itself runs on Microsoft's cloud, not your machine. Their docs are in mild disagreement right now about whether the AutoSave toggle must be on, but none of them describe Copilot working on a file that never touched Microsoft's cloud. The export sitting on your desktop, the one you downloaded from Stripe precisely so you could look at it quietly, is exactly the case Copilot is not built around. The supported formats for its analysis are the Excel workbook family, too, so a raw CSV gets converted and uploaded on its way in.

Then there is the question of which Copilot you actually have. On work accounts it is a paid add-on per user, per month, on top of the Microsoft 365 plan you already pay for, and it is priced for a company rolling it out rather than one person with a question. On home subscriptions it now comes included, but metered: a fixed monthly allowance of AI credits, spent by exactly the kind of data analysis this page is about, and available only to the subscription owner on a Family plan. Whether a given question is covered depends on which of these you hold, which is a strange thing to have to know before asking about a spreadsheet.

Last, the small print that bites first: Copilot's analysis wants your data shaped as a table or a table-like range, and Microsoft says plainly that unstructured data is not supported. Real exports are rarely tidy, and Copilot expects you to have done the tidying.

How does Paperswift approach the same job?

Paperswift starts from the file you actually have. You open the CSV or Excel export in your browser (no conversion, no move to OneDrive, no tenant) and ask in plain English. The query runs on your own machine, and what comes back is a chart built for the answer with the query readable underneath, so you can check what it counted. Once a question is worked out you can save it and rerun it against next month's export, and the rerun involves no AI at all. How it works has the longer version.

It is a narrower tool than Copilot. It will not write formulas into your workbook, reformat your sheet, or edit anything in place; the spreadsheet stays exactly as it was. What it does is answer questions about the data, privately, and show its working.

Where does each one hit a ceiling?

Excel has a hard ceiling and a soft one, and the soft one arrives first. The hard ceiling: a worksheet holds a little over a million rows, and a larger export simply does not open in full. The soft one shows up from a few hundred thousand rows, when scrolling, recalculating, and filtering all start to drag as the grid grows. Copilot works on the same grid, so anything that makes Excel labour makes its assistant labour with it — the sheet is both the interface and the bottleneck.

Paperswift is not built on a grid, which is why the same file sizes behave differently. The export is read by a columnar query engine inside your browser, made for exactly this kind of scanning and grouping, with no million cells to draw. Questions on a file with millions of rows come back about as quickly as on a small one — the working range on the paid tier is around five million rows. The practical bound is your machine's memory, not a worksheet ceiling.

What does each one cost to set up and run?

Copilot's cost depends on who you are. A company buys it as a per-user add-on on a yearly commitment, which makes sense at the scale of a rollout and rarely at the scale of one person. A home user gets it inside a Microsoft 365 subscription, where the real currency is the monthly credit allowance rather than the sticker price — the plans page lists the figures, the credits page the limits. Either way you are not buying a data tool; you are buying an AI layer across Office that also does data. Current numbers are on Microsoft's pricing pages.

Paperswift has one job and mostly no meter. Questions are unlimited on every tier, including the free one, and that is architectural rather than generous: the model only ever receives your column names and their types, so a large file costs no more to ask about than a small one. The free tier answers as many questions as you like and shows every chart on screen; taking results back out (downloading, saving a chart, exporting) is what the paid tier adds. Current numbers are on our pricing page.

What actually leaves your computer?

With Copilot, the workbook lives in OneDrive or SharePoint, and when you ask a question the prompt and the relevant workbook content are processed on Microsoft's servers. Microsoft's commitments here are serious — the data is encrypted, it is not used to train foundation models, and business tenants sit under contractual terms. If your organisation has already accepted those terms for its documents, Copilot adds little new exposure.

Paperswift's answer is structural rather than contractual. The rows never leave your machine, because the analysis happens there. The one thing the AI uses when you ask a question is your column names and whether each column holds text, numbers, or dates — never the values in any row. There is no tenancy to trust because there is no upload. The privacy page sets out exactly where that line falls.

Which one should you choose?

Choose Copilot when your files already live in a licensed Microsoft 365 tenant, the data fits comfortably in a worksheet, and the work is Excel work: formulas, formatting, a chart dropped into the sheet you are about to present. It is the obvious choice there, and switching tools to avoid it would be effort spent against yourself.

Choose Paperswift when the file is an export on your machine, the job is a question rather than a spreadsheet edit, and it matters that the rows stay where they are — or when there are simply too many of them, because an export a few hundred thousand rows deep is already past the point where Excel feels like Excel, and a million-row one will not open in full at all. That case is common; most exports from Stripe, HubSpot, or QuickBooks begin life as a local download, and it is the case Paperswift was built for.