Skip to main content
VUONG PHAM, home
← All Insights

ArticleDocument workflows

What “offline” should mean for a document tool

Many tools call themselves private. For sensitive files, check where processing runs, what the network sees and what is kept afterwards.

Vương Phạm · Published · 4 min read

Contracts, tender packages and identity documents are often covered by rules that forbid uploading them to third-party services. When someone needs to split, merge or OCR such a file, “private” on a product page is not enough. The useful question is where the work actually happens.

Three questions to ask

  • Where does processing run? On the device, or on a server after upload?
  • What does the network see? Does the tool send file content, page images or extracted text anywhere, including for OCR or analytics?
  • What is kept? Are temporary files, caches or thumbnails left behind after the job?

Note

You can check the second question yourself: disconnect from the network and run the task. A tool that is offline by architecture will still work.

Why OCR is the hard part

Splitting and merging are straightforward to run locally. OCR is where many tools quietly send pages to a cloud service, because good recognition models are large. Vietnamese adds diacritics that generic models often drop or confuse, so quality must be measured on real Vietnamese documents, not assumed.

What to expect from a trustworthy tool

  • A plain statement of network behaviour that you can verify.
  • Published OCR accuracy on stated sample documents.
  • Clear limitations, such as handwriting or low-resolution scans.