Comparison
Browser-only vs. server-based anonymization
Four ways to remove names and identifiers from a document before it reaches an AI, compared on the questions that matter: where the file is processed, what it costs, what survives, and what you can verify yourself. No product names; only the categories.
Side by side
What each approach does with your document
Read the rows as questions to ask any tool. The Unformation column describes exactly what this site does today; the other columns describe the category in general, so individual products may differ.
| Side by side | Unformation | Server-based online anonymizers | Manual find/replace | Enterprise DLP / redaction software |
|---|---|---|---|---|
| Where the document is processed | In your browser tab, in a Web Worker. 0 network requests during processing. | On the provider's servers; the file is transmitted, processed and usually stored for a while. | On your device, inside your editor. | On your device or on company servers, depending on the product and deployment. |
| Upload required | No. The file never leaves your device. | Yes. | No. | Sometimes; agent-based tools scan locally, cloud tools upload. |
| Works offline | Yes, after the first page load. Switch Wi‑Fi off and it keeps working. | No. | Yes. | Usually yes for endpoint agents; no for cloud consoles. |
| Account / cookies | No account, no cookies. | Usually an account; often cookies and tracking. | None. | Corporate identity, licences, admin console. |
| Price | Free. | Free tier with limits, then a subscription. | Free (you already own the editor). | Per-seat licences; typically procurement and onboarding. |
| DOCX / PPTX / XLSX formatting preserved | Yes. Text is replaced inside the original XML runs; styles, tables, notes, headers and images stay. | Varies; many return plain text or a rebuilt file. | Yes, if you edit in the native application. | Usually yes for supported formats. |
| PDF handling | Text-layer PDFs: comes back as DOCX + Markdown. Scanned PDFs are not supported (no OCR). | Often includes OCR; the scan is uploaded for it. | PDFs cannot be reliably edited by hand. | Usually strong: OCR, visual redaction, burn-in. |
| Consistent pseudonyms + reversible mapping | Yes. Person 1 stays Person 1 across files; a mapping table (CSV/JSON) restores names in the AI's answer. | Sometimes; reversal usually needs their service again. | No. Each replacement is manual and easy to miss. | Sometimes; often redaction only (irreversible). |
| Detection layers | Patterns with checksums (IBAN, ID, card, email, phone), your dictionary, and an optional named-entity model that runs in the browser. | Server-side models; you cannot inspect what runs. | Only what you search for. | Extensive rule libraries and classifiers, tuned by admins. |
| Verifiable (Network tab) | Yes. Open DevTools and watch: no request carries your document. | No. You must trust the privacy policy. | Yes. | Partly; depends on the vendor's architecture. |
| Best for / limits | Documents you are about to paste into an AI; quick, private, no setup. Limits: no OCR, no images, up to 10 files of 50 MB. | Occasional use when you can accept an upload; large-scale OCR. | A single short document with a handful of known names. | Policy enforcement across an organisation; audit trails; compliance mandates. |
When to use which
An honest recommendation
You are about to paste a document into an AI
Use a browser-only tool. The document never leaves your device, formatting survives, and the mapping table lets you restore names in the answer. This is exactly what Unformation is built for.
Your PDF is a scan or a photo
You need OCR, which we do not have. Either use an OCR step first (offline OCR software or an enterprise tool) and then anonymize the resulting text here, or accept the upload that a server-based OCR service requires.
You must enforce a policy across a team
Enterprise DLP or redaction software is the right category: central rules, audit logs, integration with mail and storage. Unformation is a per-document tool with no admin console.
It is a quick one-off with two or three names
Find/replace in your editor is fine. Be careful with suffixes, plurals and possessives; that is where manual replacement usually misses one.
You handle many documents every week
A browser-only tool with a dictionary and consistent labels saves the repetitive part and keeps the files private. If you also need automation across systems, look at DLP.
Questions
Comparison questions
Is a browser-only tool really as private as it claims?
You can check instead of trusting: open the browser's Network tab, anonymize a document and confirm that no request carries the file. Unformation makes 0 network requests during processing; the only optional download is the detection model, once.
Why would anyone use a server-based anonymizer then?
OCR for scanned documents, very large batches, and integrations. Those need compute and storage on a server. If your document is already text, none of that is required.
Is manual find/replace good enough?
For a short document with a few known names, yes. It fails on the things you did not think of: emails, IBANs, IDs, a name with a Turkish suffix, the second spelling of a company. Automated detection lists those for you.
Does Unformation replace enterprise DLP?
No. DLP enforces rules for an entire organisation and produces audit trails. Unformation helps one person anonymize one document before an AI sees it. They can coexist.
What about accuracy compared with server-side models?
Patterns with checksums are exact. Name detection depends on the optional model; the multilingual model is comparable to common server-side NER models but not perfect, which is why the review step exists. Always read the list before you download.
Try the browser-only approach
Drop a document, watch the Network tab stay quiet, download the result.