Why convert documents locally instead of uploading them
Converting documents in your own browser means contracts, medical records and internal files never leave your machine - which removes an entire category of risk that no vendor promise can.
The documents people want to convert are the sensitive ones
Nobody builds RAG over public information; they build it over contracts, patient notes, board papers and internal wikis. That is precisely the material an upload-based tool asks you to send to a third party, and precisely the material most organisations cannot send.
A promise is not a control
A privacy policy says what a company intends. Local conversion changes what is possible: if the bytes never leave the tab, there is no copy to be breached, subpoenaed, retained past its policy, or used for training. That distinction survives changes of ownership and terms of service.
What it costs
Your machine does the work, so a large corpus takes longer than it would on a server, and very large models are impractical to download. Local conversion is excellent for text extraction and chunking, and genuinely limited for anything needing a large model - OCR and speech recognition especially.
A reasonable split
Convert locally by default. Move to a server for the specific jobs that need real horsepower, with eyes open about what you are sending and why. The choice is per-document, not per-tool.