Anoni alternative: what the free tool offers, and when to choose AnonymFiles

What Anoni offers

Anoni presents itself as a desktop app for “local anonymization for AI”, designed and published in France (our translation). Its approach is close to AnonymFiles': you drop a document, the app finds sensitive data on your computer, you check it, then you give the copy to your usual AI and get the real names back in its answer.

• Free. According to its FAQ, all current features are free, with no subscription; a Pro edition is “in development”, with no published price.
• Three systems. macOS 13.4 or later (macOS 14 for AI detection on Apple silicon Macs), Windows 10 and 11, 64-bit Linux; the app is also on the Mac App Store.
• Formats. PDF, DOCX, Excel, email (.eml), TXT, Markdown, CSV and images, with local OCR in French and English for scanned PDFs.
• Two modes. Pseudonymization replaces “Jean Dupont” with a realistic name such as “Marc Leroy”, reversible thanks to an AES-256-GCM encrypted vault; redaction applies permanent black bars in the exported PDF.
• Layout. Its changelog states that since August 2026 Word, Excel, PDF and scans keep their format, and that decoding the AI's answer returns a Word document.
• A published benchmark. On 210 synthetic French legal documents (June 18, 2026), Anoni reports 94.8% strict recall and 99.0% of sensitive characters covered, and 92.7% strict recall on 500 documents from an independent set (ai4privacy OpenPII). That is a broader benchmark than the one AnonymFiles publishes today.
• Extra tools. An anonymization report with a SHA-256 fingerprint, a watermark repeated on every page, a local MCP connector for AI agents, masking of signatures and stamps.

Three points matter for a firm, though. On the network side, Anoni downloads its engine and models at first launch, asks for an email address when downloaded from its website, and by default reports one launch per day, without content (“Signaler mes ouvertures” turns it off, according to its changelog); updates install automatically. In the free version, exports carry an “anoni.dev” watermark. Finally, processing a whole folder with the same pseudonyms from one document to the next is presented in its changelog as a Pro edition feature, whose page says the scope “is still being defined” (our translation).

Anoni vs AnonymFiles: eight criteria

Here are both tools on the eight criteria of our comparison, which follow what a firm needs to hand a case file to an AI without breaching secrecy. Anoni has changed a lot since our October 12 review: its score rises from 4.5 to 6 out of 8 based on what its website states today. The license check at launch and the daily launch signal, on by default, cost it the “no network call at all” criterion, as for Lexanon, Tacitum, Hexagone AI and Occlira.

✓: yes. ◐: partly. ✗: no. n.c.: not disclosed by the publisher, scored zero. One point per criterion met, half a point when partly met, as in our comparison of anonymization software. For AnonymFiles on Windows, the detection models are downloaded once, on first processing, without any document text; on Mac, they are included.

When Anoni is still a good choice

Anoni is a good option if you want a free tool for the occasional document, if you work on Linux, or if you prefer realistic pseudonyms (“Marc Leroy”) to placeholders (“[PERSONNE_1]”). Its detection benchmark, published on a broader set, is also a serious argument. If the “anoni.dev” watermark does not bother you and the launch signal, once turned off, suits you, there is no reason to pay for occasional use.

A note on realistic pseudonyms: they make the text more natural for the AI, but a reader can no longer see what was replaced. A placeholder like [PERSONNE_1], on the other hand, shows at every review that a piece of data was removed. It is a choice of method, not a flaw.

Why AnonymFiles for a firm

AnonymFiles, a French document anonymization software, prepares a redacted copy of a whole case file on your computer before you hand it to ChatGPT, Claude, Copilot or Le Chat. Each person keeps the same placeholder in every document: Jean Martin becomes [PERSONNE_1] in the pleadings and in the emails alike. Scanned PDFs and images are read by local OCR and redacted in the image; a Word file comes out as a .docx redacted in place, layout preserved; .eml and .msg emails, .xlsx and .csv spreadsheets and .pptx presentations are supported. The original and the copy are shown side by side, and nothing is exported without document-by-document approval. After the AI, the app puts the real names back into its answer, as text, Word or PDF; the mapping table stays on the computer, encrypted (AES-256-GCM).

Network-wise: no account, no telemetry, a license checked on the computer with no network call; on Mac, the detection models are included and nothing connects. Details are on the security and network scope page, and the published detection benchmark explains how to check it yourself.

Price: €290 excl. VAT per seat per year for one or two seats, €240 excl. VAT per seat from three seats, a quote above twenty, with no page quota; 14-day trial. AnonymFiles is coming soon for Apple silicon Macs and Windows. It is built for professions bound by secrecy: document anonymization software for lawyers, for notaries, for chartered accountants, and the other profession pages below.

Compared with Anoni, the difference comes down to four verifiable points: no network call at all, not for an email address, a launch signal or the license; the whole case file processed with the same placeholders without waiting for a paid edition; Outlook .msg emails and .pptx presentations; copies with no publisher watermark, ready to send as they are.

Other Anoni alternatives

• Our comparison of anonymization software: ten tools scored on the same eight criteria, with the cost for a five-seat firm.
• Lexanon alternative and Hexagone AI alternative: the two French tools sold to firms on a yearly subscription.
• Tacitum alternative: the Windows tool sold for €79 once.
• Adobe Acrobat redaction alternative: when the Redact tool is no longer enough.