OCR that never leaves your servers

Scanned documents in Polybase are now searchable. A local OCR engine reads scanned PDFs and images uploaded to the workspace and extracts their text — and it does so entirely on the server your workspace runs on. No page is sent to an external service; the documents never leave your infrastructure.

Scans are usually where searchability ends. The signed contract, the incoming invoice, the archived paper file — the documents teams most need to find again are precisely the ones that arrive as pictures of text rather than text. Most products close that gap by shipping the pages to a cloud OCR API, which is exactly the wrong move for the documents least suited to travel.

Polybase closes it locally instead. Once a scan is processed, its text feeds the same full-text search as everything else in the workspace: searching for a clause or an amount now also surfaces the scanned page that contains it. The recognised text is available in the document viewers too, so text in a scan can be found and worked with rather than merely looked at.

This is what the sovereignty promise looks like when it gets concrete. "Your servers, your keys" is easy to say about chat messages; it has to hold for the hard cases — the signed, the stamped, the confidential. Because the engine is part of the deployment itself, it works wherever Polybase works, including self-hosted and air-gapped installations with no outbound network at all.

Local OCR is live now for scanned documents across the workspace. For the wider picture of how Polybase handles your data, see our security overview.