English
AI

Privacy: what leaves your computer when you use AI

Updated 2026-09-04 · 2 min read · For desktop macOS 0.0.9 / Windows 0.0.7

Without AI, nothing leaves. With it, that one request goes from your computer straight to the provider you configured — your question, the relevant column names and a few sample rows, or the image itself. This page says exactly what each entry point sends.


The principle

The desktop app's engine and interface both run on your machine. Data goes out only when you actively use an AI feature, and it goes to the provider you filled in yourself, never through a server of ours — there is no server of ours on that path at all.

What happens to it after that is between you and your provider.

What each entry point sends

Entry point What goes out
Ask AI in a panel Headers (column names) plus a few sample rows
AI formula Your description, plus the names and sample rows of the columns involved
AI Generate new row Your description, plus headers and sample rows
Ask AI (chat) The text you type, the full contents of files you upload, your search terms
Convert Image to Table The image itself
Dashboard AI Your description, plus headers and sample rows of the table you picked
AI functions (ai.generate and friends) The relevant columns of every row
Web functions (web.crawl and friends) The address or keyword you wrote as an argument, sent to that site

"Sample rows" means a handful of rows so the model can see what the data looks like, not the whole table. AI functions are the exception: they work row by row, so the whole column is sent, one row at a time.

What never goes out

  • Your API key lives in settings.json on this machine. It is not backed up and not uploaded.
  • The table data itself sits in a local folder. There is no sync of any kind.
  • Error reporting (Desktop settings → Privacy & diagnostics), when enabled, sends anonymous crash and exception information — no cell contents, no file contents, no personal data. You can switch it off at any time.

How to work with this

  • For data that cannot go to a third party — signed contracts, identity documents, unreleased financials — simply don't use AI on that table. Everything else keeps working.
  • If your team requires a specific model service (an internally hosted compatible endpoint, say), put its address in Chat completions endpoint and requests go only there.
  • To stay completely offline: don't configure a key. The rest of the app is fully usable without one.
Did this page not answer your question? Browse the help centre, or write to support@tabledi.com — the inbox is read by the people who build TableDI.