For institutions
unPaper lets any organization get frontier-quality documents from whichever AI it is allowed to use — a company-approved assistant, a European model, or an open-weight model on its own hardware — with no document content uploaded, no AI inside the tools, and every artifact a self-contained open file that prints perfectly offline and converts to fully editable Office formats on the user’s own machine.
Sovereignty here is not a promise in a contract — it is a property of the architecture. Every claim below is checkable in minutes. We would rather you test than trust.
01
The tools are static pages and local command-line programs. There is no document upload, no account, and no telemetry on content. The hosted site measures only cookie-free aggregate page traffic, as disclosed in the privacy policy.
Check it
Open unpaper.ai, disconnect the network, and use both tools. They work with the Wi-Fi off.
02
unPaper contains no model and calls no AI API. It gives your AI a strict format plus instructions, and deterministically converts the result. Change vendors, use a European model, run one locally: the tools neither know nor care.
Check it
The format works pasted into any current assistant — including open-weight models.
03
The file format (unPaper HTML) is publicly specified with conformance levels, a free validator, and an open pattern library. Documents are ordinary HTML that any browser on Earth renders — today and in forty years. PowerPoint is an export, not the source of truth.
Check it
The spec at /spec, the validator at /check, and the public repository.
04
Conversion fidelity is scored by a harness; model output quality is scored on a published rubric. When we say a weaker or local model produces frontier-grade documents under this contract, that is a measurement you can reproduce.
Check it
Every public number traces to a published test artifact.
Document content never reaches a server — not unPaper's, not any third party's: every tool processes it client-side, on the user's own device, by design. The hosted site produces ordinary access logs and cookie-free aggregate page analytics, disclosed in the privacy policy, but no prompt or file enters that data flow. Document processing therefore needs no cloud transfer or document-processing DPA.
Static site or offline CLI; no accounts to manage; works inside locked-down environments with the assistant you already approved. Nothing to integrate — and nothing that competes with your existing suite choices: unPaper emits open files and exports to Office formats.
Converted documents carry named objects and alternative text generated from the document's own semantics — born accessible rather than remediated, aligned with EN 301 549 obligations.
Every document is one self-contained file, no external fonts or scripts, printing identically offline — closer to the spirit of PDF/A than most PDFs, and readable without permission from any company, indefinitely.
unPaper makes the models you are allowed to use more capable. With a strict contract and a deterministic repair loop, open European and local models produce documents competitive with the frontier's — we publish the measurements. Capability, without new dependencies.
“Document content never reaches a server — ours or anyone else’s. It is processed entirely on the user’s device.”
We ask nothing but use. The hosted tools are free, no login exists, and document content never leaves the device. We are interested in feedback from real institutional workflows — briefing notes, status reports, decks.
hello@unpaper.aiunPaper is built and stewarded by reHeritage GmbH, Berlin (HRB 254687 B) — a European company maintaining the open standard and offering commercial licences for embedding, enterprise deployment (including air-gapped), and supported rollout with your company template, for organizations that need them. Use stays free — including commercial use; the format and the validator are free and stay free.