Veilgate is a local prototype for filtering private information from school email before sharing selected text with an online assistant.
The workflow combines on-device OCR for printed text in image attachments, local rules for obvious identifiers and secrets, and a local language model to identify other private passages. Users paste a message or open an .eml file, review the filtered text, and choose what to share.
The tool is experimental and does not provide a general privacy guarantee. See the GitHub repository for its current scope, supported inputs, and evaluation limitations.