Offline tools · ranked and sourced
The 10 best offline anonymization tools in 2026
No single tool is best at everything, because these tools do two different jobs.
Audio, video and every other kind of file
CaseGuard Studio
The only offline suite here that redacts body-cam footage and call recordings as well as documents, and the only one to score 10 on formats. It ranks ninth overall because this ranking counts price: five users on the plan that covers documents, audio and video together cost US$22,740 a year.
Text, including text in scans and images
promptShield
Ranks first at 8.2 out of 10, on detection, workflow and languages. Four tools follow at 7.7: Redactwizard, built for air-gapped networks, and Philter Desktop, Quieta and CamoText, the strongest low-cost alternatives.
Anonymization software finds personal data in a file and removes it or replaces it with something harmless. That data includes names, addresses, ID numbers, account numbers and health details. This ranking covers only offline tools: the documents are processed on your own computer, never uploaded to the vendor. Most law firms, hospitals and HR departments require that, and so does anyone who wants to use AI on confidential files without handing them to a third party.
We make promptShield, which is one of the ten. Every tool, ours included, was scored by the same published rules, and every fact about a competitor links to its source.
Key facts
Facts checked on 31 October 202635
tools reviewed
13
scored
10
ranked
- Ranked tools must process documents on your machine and find personal data automatically.
- Two scores per tool: technical (60% of the overall score) and value for money (40%).
- The detection score rates how each tool's detection is built and documented, on a published scale. It does not measure how much each tool catches: we only measured our own.
The ranking
The ranking at a glance
- 01
promptShield
Using AI on confidential documents
Windows, browser · about US$1,500
Technical
8.6Value
7.58.2Overall
- 02
Redactwizard
Air-gapped networks, medical files
Windows 11 · about US$1,560, one-time
Technical
8.2Value
7.07.7Overall
- 03
Philter Desktop
Low-cost redaction on Windows
Windows · US$995
Technical
7.2Value
8.57.7Overall
- 04
Quieta
Masking files before AI on a Mac
Windows, Mac · about US$840
Technical
7.1Value
8.57.7Overall
- 05
CamoText
Strictly no network, text documents
Windows, Mac · US$445, one-time
Technical
6.9Value
9.07.7Overall
- 06
Anonimizer
Polish users
Windows, Mac · about US$1,020
Technical
6.7Value
7.57.0Overall
- 07
AiNonymizer
Smallest budgets
Windows, Mac, Linux · about US$225
Technical
5.5Value
9.06.9Overall
- 08
Stript
German and English legal work
Windows, Mac, Linux · about US$2,630
Technical
7.0Value
6.06.6Overall
- 09
CaseGuard StudioSuite
Audio and video as well as documents
Windows · US$22,740
Technical
8.2Value
4.06.5Overall
- 10
Hexagone AI
An AI workspace in the same app
Windows, Mac · about US$5,310
Technical
7.4Value
5.06.4Overall
Price: five users, first year, list price. Ties are ranked by the technical score.
The map
Technical score against value for money
promptShield and Redactwizard are the only tools with a technical score of 8 or more and a value score of 7 or more; promptShield is higher on both.
Method
How we rated
Each tool gets two scores out of 10 and a breadth label. The overall score is 60% technical and 40% value for money. Ties are ranked by the technical score.
The weights fit one reader: a professional who has to remove personal data from documents, often before sending them to an AI tool.
Technical score
How the tool finds personal data on its own, and what evidence the vendor publishes about it. Scored on the scale below.
Whether detection runs fully on the device, whether an account is required, what the app sends home (licence checks, telemetry, cloud features), and whether it works on an air-gapped network.
Six features, scored on the scale below.
PDF, Office files, scans (OCR), email, images, audio, video; Windows, Mac, Linux.
Languages detected natively, and country-specific IDs such as IBANs or national ID numbers. A tool that publishes no language list scores 3.
Detection scale
We did not test the other nine tools on the same documents, so their detection score cannot say how much each one catches. It says how detection is built and how much the vendor shows about it.
- 9Local AI detection, plus published accuracy figures across several languages
- 8Local AI detection with broad documented coverage, such as specialist models, but no accuracy figures
- 7A named AI model combined with patterns or checksums, or accuracy figures for one or two languages only
- 6Automatic detection with the data types listed, but the method not disclosed
- 5Automatic detection, with neither the method nor the data types disclosed
- 4Editable rules, documented in detail for one country only
- 3Pattern or word-list search only, with no automatic name detection
Two limits apply. Accuracy figures published by a vendor, ours included, come from tests the vendor chose and ran. And a tool whose method sounds weaker on paper may catch more in practice. promptShield's 9 rests on our own published benchmark: the documents, labels and scorer are public, so anyone can rerun it, but nobody independent has yet. Without those figures it would score 8 on this scale.
Workflow scale
Every tool starts at 4 and gains one point for each of these features that the vendor documents for the plan we price. A tool with all six scores 10, one with none scores 4. A feature that is announced but not shipped, or sold only on a plan we did not price, does not count.
+1
Review before export
You see what was found and can correct it before the output is written or used.
+1
Add what was missed
You can mark by hand, on the document, something detection did not find. A word list or rule that is searched for does not count.
+1
Batch processing
Several files or a whole folder in one action.
+1
Audit trail
A saved record or report of what was changed.
+1
Team management
An administrator manages or oversees the team's users or licences in one place.
+1
Reversible
The original values can be restored after the AI step.
| Tool (plan) | Review | Add by hand | Batch | Audit trail | Team | Reversible | Workflow |
|---|---|---|---|---|---|---|---|
| promptShieldPro | 10 | ||||||
| AnonimizerTeam | 10 | ||||||
| CaseGuard StudioUltimate | 9 | ||||||
| RedactwizardPro | 8 | ||||||
| Philter Desktop | 8 | ||||||
| QuietaPro | not documented | word list only | 8 | ||||
| CamoTextInternational | 8 | ||||||
| StriptPro | rules only | announced | 8 | ||||
| Hexagone AIPro | not documented | Enterprise only | 8 | ||||
| AiNonymizerTeam | rules only | 7 |
One team feature is not scored: a compliance monitor showing who on the team uses the tool and when. Only promptShield documents one, and a criterion that only one tool meets would measure nothing but that tool.
Value for money
Value for money follows a formula. We take what five users pay in the first year at list price, and divide it by the technical score. The fewer dollars per technical point, the higher the value score.
- The plan we scored. We price the cheapest plan that includes every feature counted in the technical score and has no stated volume limit, billed annually where the vendor offers that. Where a cheaper plan exists, the review names it and what it leaves out.
- No promotions. Launch offers and first-year discounts are left out, because they end. Where one exists, the review gives the discounted figure and what it would do to the score.
- Prices as published. Hexagone AI's and Anonimizer's prices exclude VAT and Stript's include it; without VAT, Stript's score would not change. Redactwizard's shop does not say. Euro and zloty prices are converted at about US$1.12 per euro and 3.88 zloty per dollar.
9
under 75
US$ / point
8
75 to 150
US$ / point
7
150 to 250
US$ / point
6
250 to 500
US$ / point
5
500 to 1,000
US$ / point
4
over 1,000
US$ / point
A free plan you can keep using, not just a time-limited trial, adds half a point.
Breadth
Built for one job: anonymizing documents.
A broad redaction platform covering documents, images, audio and video.
A missing fact scores low rather than being guessed.
1. promptShield: best for documents you want to use with AI
SpecialistWindows, browserWebsite
Overall
8.2
Technical
Value
promptShield is a desktop app that finds personal data in documents and either removes it for good or replaces it with codes you can decode later. Detection runs on your computer. It is built for one job: making a confidential document safe to send to ChatGPT, Claude or Gemini, then putting the real values back into the answer.
Strengths
- Measured detection. On 48 test documents in 8 languages, it left 92 of 4,489 sensitive words readable (97.9% caught), with 1.9 wrong highlights per document. The documents, labels and scorer are public, so anyone can rerun the test.
- Nothing leaves the machine. Detection runs locally, including OCR on scans. The free trial needs no account.
- Two modes. Redact removes data permanently. Encode swaps each value for a code such as
[P0ABCD]. The same name gets the same code across every file in a folder, so the AI can follow who is who, and Decode restores the originals in its answer. - 8 languages, 17 countries. Native detection in English, French, German, Spanish, Italian, Dutch, Portuguese and Polish, with checksum validation for national IDs and IBANs.
- Review, and add what was missed. You check and correct every finding before download, and draw a box over anything detection missed. A drawn region can be repeated on every page, or saved as a template for documents with the same layout, which covers stamps, barcodes and signatures that hold no text. Many files can be processed at once.
- Compliance monitoring. It is the only tool here whose documentation describes a team compliance monitor: who has used the app and when, members flagged after a period of inactivity you choose, a team-wide audit log naming the member and the machine, and dated reports for auditors. The log holds counts and dates, never document content. A license manager assigns and reassigns seats by email.
Weaknesses
- Windows only on the desktop. A Mac version is not out yet; there is a waitlist. Mac users can use the free browser version, which also processes files locally.
- Documents only. No audio or video redaction.
- Licence check-ins. The app re-checks its licence online regularly, so a machine kept permanently offline needs the Pro plan's offline licence route.
- Not the cheapest. AiNonymizer, CamoText, Quieta, Philter and Anonimizer cost less per technical point.
- Young vendor. Few public reviews so far, and the benchmark was run by us, even though it is published.
Pricefive users, year one: about US$1,500
Free browser app and a 7-day free trial with no account. Week Pass US$9. Basic from US$24 a month, Pro from US$47 a month (billed annually). A five-person Pro team costs about US$1,500 a year. We score Pro, because Basic has no license manager, audit log or compliance monitor.
Best for
Law firms, accountants, HR and healthcare teams who want to use AI on real documents without sending personal data to it.
Try it freeSourcesPresidio benchmarkfeaturescompliance monitor demopricing
2. Redactwizard: best for air-gapped networks and medical files
SpecialistWindows 11Website
Overall
7.7
Technical
Value
Redactwizard, from Germany, is a portable Windows 11 app designed for air-gapped installations. Its local AI models suggest redactions for review or apply them automatically, and the Pro edition adds medical and EU legal models.
Strengths
- Designed for air-gapped networks. Local OCR and built-in AI work without a network connection, according to the vendor.
- Specialist models. Pro has four local models: general personal data, a helper model, medical, and EU legal.
- EU languages. Its personal-data model is focused on the 24 official EU languages.
- True PDF redaction. Pro produces native vector PDF redaction; both editions can output rasterized redactions. Local OCR includes handwriting.
- Review and batch. Detected matches arrive as suggestions you accept or reject, you can draw a box or select text for anything missed, and whole folders are processed in one operation. An audit log records every action and exports as JSON, PDF or CSV.
Weaknesses
- Windows 11 only.
- No reversible mode documented, so there is no way to send a document to an AI and put the real values back in its answer.
- No team management. Each activation code covers one workstation, extra users cost €199 each, and the audit log stays on each machine.
- Very new. Founded in 2026, with no customers named and no legal company name on the page.
- Price varies by source. €399 (Legal) or €599 (Pro) on the vendor's site; directories list other figures.
Pricefive users, year one: about US$1,560, one-time
One-time and perpetual. Pro €599 for the first user and €199 for each additional user, so five users cost €1,395 (about US$1,560). Legal €399, then €99 per additional user. The shop does not say whether VAT is included. We score Pro, because the medical and EU legal models and vector redaction are Pro only.
Best for
Hospitals, public bodies and law firms whose machines are kept off the internet.
SourcesRedactwizarddocumentationshopLegal Technology Hub listingCapterra listing
3. Philter Desktop: best low-cost option on Windows
SpecialistWindowsWebsite
Overall
7.7
Technical
Value
Philter comes from Philterd, which has built the open-source Phileas redaction engine since 2019. Philter Desktop is a Windows app that works offline. Its only network call is a manual update check.
Strengths
- Offline by design. Processes files on the machine, with no network calls beyond the update check you trigger.
- Many output modes. It can redact, mask, encrypt, shift dates, or swap in realistic substitute values, kept consistent within a context.
- Solid detection stack. A GLiNER language model, name dictionaries, and patterns with checksum validation (Luhn, mod-97).
- Wide office formats. PDF, DOCX, XLSX, CSV, TXT, RTF, EML and MSG.
- Careful output. You preview and adjust what will be redacted before saving, and can draw a box or select text to redact what it missed. It can re-scan the finished file, and it writes a report of every change. Watched folders and a command line handle batches.
- Open-source core. Phileas is Apache-2.0, so the engine can be inspected.
Weaknesses
- PDF handling. Desktop flattens PDFs to images, and its OCR on scans is described as "best-effort".
- No published language list. We found no list of supported languages or data types.
- Small community. About 110 GitHub stars for Phileas, so fewer people test it in public.
- One-way and single-user. No way to restore the original values after an AI step, and no team management.
- Windows only.
Pricefive users, year one: US$995
Free for personal use and evaluation. Commercial use costs US$199 per user per year. The first year is currently offered at US$79.60 per user, with no end date stated; we score the regular price. At the first-year price, five users would cost US$398, and Philter's overall score would be 8.1, still second.
Best for
A small office that wants an offline redactor at a low price and can live with flattened PDFs.
SourcesPhilter Desktopredacting documentsPhilterd pricingPhileas on GitHubcompany
4. Quieta: best for masking files before AI on a Mac
SpecialistWindows, MacWebsite
Overall
7.7
Technical
Value
Quieta, from Paris, is a Mac and Windows app that masks personal data with a local AI model (about 1 GB), then restores it after you have used an AI tool. A "Mirror" feature keeps masked copies of whole folders for AI agents to read.
Strengths
- Local model. "Your files never upload", and the vendor says it keeps working with Wi-Fi unplugged.
- Reversible tokens, with masked folder copies for AI agents.
- Formats. PDF, DOCX, TXT, XLSX, CSV and OCR on Plus; Word, Excel and PowerPoint, batch processing and custom patterns on Pro.
- Team tools on Pro. Audit reports in CSV or PDF, and an admin dashboard listed as showing team usage. The pages do not say what it shows, and the privacy policy says no usage statistics are collected.
- Free plan of 50 masks a month, no account needed.
Weaknesses
- Masking, not redaction. No permanent-redaction export is described, so it does not produce a document you would release.
- No review step or hand-marking documented. The pages we read describe masking and restoring, not checking what was found before use, or marking what it missed. Allow and block lists are the documented controls.
- Language claim unverified. The vendor says "all languages" without a list.
- Network behaviour not described. Licence keys come by email; what the app sends home is not documented.
- Little company information. No legal entity is stated on the pages we read.
Pricefive users, year one: about US$840
Free (50 masks a month), Plus €70 a year, Pro €150 a year, both with unlimited masks. We score Pro, because Plus has no batch processing or audit reports. A lifetime Pro licence for version 2 costs €289 (a launch price of €189 ended on 31 October 2026). VAT is handled by the payment provider according to your country. 7-day Pro trial, and a 30-day money-back guarantee.
Best for
Mac or Windows users who mask files before using ChatGPT or Claude and put the real values back in the answers afterwards.
5. CamoText: best if the machine must never connect to a network
SpecialistWindows, MacWebsite
Overall
7.7
Technical
Value
CamoText, from CamoText LLC, is a Windows and Mac app built by a practising attorney. It anonymizes text documents with bundled local models and lets you put the original values back later with keys kept on your machine.
Strengths
- The strictest offline statement in this list. "No outbound or inbound network connections", no telemetry, and fully functional in air-gapped environments, according to the vendor. The trial needs no account.
- Reversible. Anonymize, then reinsert or de-anonymize with local keys.
- Review and batch. You review the findings, revert false positives, and highlight anything missed to anonymize it everywhere it appears; single files or whole folders, plus a command-line add-on for automation.
- Price. One-time licences: US$69 (Pro), US$89 (International).
Weaknesses
- No OCR. A scanned PDF with no text layer is not supported.
- Detection method not disclosed. "Bundled local models and custom algorithms" covering 20+ data types.
- Few languages. English in Pro; English, Spanish, French, German and Italian in the International edition.
- No audit trail or team management documented.
- Little public track record. No named customers.
Pricefive users, year one: US$445, one-time
Pro US$69, International US$89, command-line add-on US$20, all one-time. 10-day free trial. We score International, because Pro detects English only.
Best for
A lawyer or consultant on Windows or Mac who works with digital documents and must be able to show the app never connects to anything.
6. Anonimizer: best for Polish users
SpecialistWindows, MacWebsite
Overall
7.0
Technical
Value
Anonimizer, from Trustlayer Sp. z o.o. in Poland, is a Windows and Mac app that works locally on your computer. It replaces personal data with labels such as [OSOBA_1] or with fictitious values, or blacks it out, and restores the real values locally after the AI step.
Strengths
- Three modes. Labels, fictitious replacement data, or permanent blackout.
- Review before use. You check what was found, correct it, and mark more by hand: draw a box on a scan or select text to add it. Then you copy the safe version.
- Keeps formatting. PDF, DOCX, ODT, XLSX and ODS keep their layout; images and scans are read with OCR on paid plans.
- Team plan. Five seats with an online admin panel, where you deactivate devices, buy seats and see how many anonymizations each device has made. Batch ("mass") anonymization and offline work are included.
- A named company, with press coverage cited on its site.
- Free plan of 5 documents a month.
Weaknesses
- Detection method and languages not disclosed, though the product is aimed at Polish users.
- Account required, and offline work is listed only on the Pro and Team plans.
- Local log only. Each anonymization is logged on the machine that made it. The admin panel shows a count per device, not who did what or when.
Pricefive users, year one: about US$1,020
Free (5 documents a month). Starter 79 zł one-time for 20 documents. Pro, one seat, 79 zł a month billed annually (99 zł monthly). Team, five seats, 329 zł a month billed annually, or 3,948 zł a year (399 zł monthly; extra seats 79 zł a month). All excluding 23% VAT. Enterprise, from 10 seats, on quote. We score the Team plan, which is also cheaper than five Pro seats.
Best for
Polish professionals who want a local tool with Polish pricing and support.
7. AiNonymizer: cheapest paid option for small teams
SpecialistWindows, Mac, LinuxWebsite
Overall
6.9
Technical
Value
AiNonymizer is a desktop app for Windows, Mac and Linux that replaces personal data with realistic fictional equivalents, then restores the originals in an AI's answer. Detection, replacement and restoration run on the computer. The privacy policy names an individual in Gubbio, Italy, as data controller.
Strengths
- Price. €49 a year for one user, €199 a year for five.
- Three operating systems. Windows, Mac and Linux, each in a "Light" and an "Advanced AI" edition.
- Realistic replacements. Plausible fake names and values, kept consistent within a document and reversible.
- Review step before replacement, and team management on the Team plans (the pages do not describe what it includes).
Weaknesses
- Detection method and languages not stated. It describes editable country-specific rules, with only Italy documented in detail.
- Licence check-ins. Activation and periodic licence validation need the internet, with a 7-day grace period offline. The trial check runs even with statistics turned off.
- No batch processing, audit log or hand-marking documented: you confirm what it found and refine its rules. No permanent redaction mode.
- Little company information. The website names no company; the privacy policy names a person, with no legal form.
Pricefive users, year one: about US$225
Pro €6 a month or €49 a year. Team 5 €199 a year, Team 10 €349 a year. A trial exists; its length is not stated.
Best for
A very small team on a tight budget, working in languages its rules cover.
8. Stript: best for German and English legal teams
SpecialistWindows, Mac, LinuxWebsite
Overall
6.6
Technical
Value
Stript is a desktop app from Munich, built by a legal-tech researcher, Dr. Ingo Glaser. It replaces personal data with consistent placeholders such as [PERSON_1], and restores the originals in the AI's reply. Detection runs on the device after a one-time model download.
Strengths
- Broad file support. PDF, DOCX, EML, MSG, CSV, RTF, ODT, XLSX, TXT and scanned PDFs, with on-device OCR.
- Windows, Mac (Apple Silicon) and Linux.
- Careful review. Confirm, reject or reclassify each finding, with a confidence threshold; a tamper-evident audit trail; batch processing on Pro.
- Clear offline rules. Pro needs one successful licence check every 30 days. Crash reports are off by default.
Weaknesses
- No hand-marking documented. You confirm, reject or reclassify findings and can add custom rules, but the pages do not describe marking a missed item on the document.
- Two languages. German and English only. Its published accuracy (F1 95.8 for German, 97.3 for English) was measured by the vendor.
- Price. €49 a month or €470 a year per user, and the free tier stops at 5 documents in total.
- Not air-gapped. New documents stop processing if the licence check lapses for 30 days, and email verification is required.
- Small business. A sole proprietorship; team features and SSO wait for an Enterprise plan that its site still announces for Q3 2026.
Pricefive users, year one: about US$2,630
Free for 5 documents. Pro €49 a month or €470 a year, VAT included. Enterprise on quote.
Best for
German or English legal teams who want reversible placeholders, including on Linux.
9. CaseGuard Studio: best generalist, and the leader for audio and video
SuiteWindowsWebsite
Overall
6.5
Technical
Value
CaseGuard Studio, from CaseGuard in Arlington, Virginia, is the broadest redaction suite in this list. It installs on your organization's own computers, and the vendor says nothing is processed or stored in the cloud. It covers documents, email, images, audio and video in one application.
Strengths
- Breadth no one else here matches. Body-cam, CCTV and dash-cam video, call recordings, PST and EML email, Office files and scans. The top plan claims 900+ file types.
- On premises. Installed locally, with air-gapped networks supported according to the vendor.
- Strong workflow. Batch processing, review before export, boxes you draw for anything missed, and audit reports, plus team collaboration tools on Ultimate. Licences are managed centrally in CaseGuard Hub, where an administrator creates users and assigns licences.
- Transcription and translation for media, which document tools do not offer.
Weaknesses
- Price. Documents and media are sold as separate plans, and only Ultimate, at US$379 per user per month billed annually, covers both. Five users on Ultimate cost US$22,740 a year. The document-only plan would cost US$16,740 for five, still the highest in this ranking.
- Metered AI. On the Doc and Media suites, AI use draws on credits, and the refund policy mentions "cloud-based AI hours". We could not confirm whether every AI feature runs offline.
- Permanent redaction only. No reversible mode for sending a document to an AI and putting the real values back in its answer.
- Roles and permissions cost extra. They are listed on Enterprise only, from 10 seats on the pricing page and from 50 in the help centre.
- Detection languages not stated. Transcription is listed as 50+ languages in one place and 100+ in another.
Pricefive users, year one: US$22,740
Per user per month, billed annually. Doc Suite US$279 (documents, email and images, no audio or video), Media Suite US$299 (audio and video, no documents), Ultimate US$379 (both, plus team collaboration and 900+ file types). Enterprise from 10 seats on quote. We score Ultimate, because the formats score counts documents, audio and video together, which only Ultimate covers.
Best for
Police forces, public bodies and law firms that release redacted video, audio and records, and have the budget.
SourcesCaseGuardpricingCaseGuard Hubmanual redactionlocal application FAQ
10. Hexagone AI: best built-in AI workspace
SpecialistWindows, MacWebsite
Overall
6.4
Technical
Value
Hexagone AI, from HEXAGONE AI SAS near Paris, is a desktop app for Mac (Apple Silicon) and Windows that masks personal data with realistic substitutes and restores them later. Its distinctive feature is a built-in AI workspace for chat, summaries and drafting, plus an overlay for Claude, ChatGPT and Cursor.
Strengths
- Published accuracy on third-party datasets. The vendor reports 97.9% recall on PII-INDEXBENCH and 95.3% on AI4Privacy across 6 languages. The vendor ran these tests.
- Consistent across a folder. The same substitute in every file of a folder, with folder-level batch processing and a local audit log exportable as CSV. Every detection appears in a preview you can edit before export.
- AI workspace and overlay for users who want the AI step inside the same app.
- Masking works offline, according to its security page and FAQ.
Weaknesses
- Not fully offline. Its terms say the licence needs an internet connection (daily during the trial), aggregated usage data is collected, and the chat features use Azure OpenAI. Air-gapped use is offered only on the Enterprise plan.
- Priced as an AI workspace. Five users on Pro cost about US$5,310 a year, the second-highest here after CaseGuard. Much of that pays for the AI chat and drafting workspace, so if you only need masking, you pay for features you may not use.
- No permanent redaction mode described, and no Linux; Mac support is Apple Silicon only.
- Fewer languages. French, English, German and Italian are named on the homepage.
- No team management below Enterprise. Pro licences are individual and tied to one device; central licence management is quoted separately.
- Hand-marking not described. The preview lets you edit what was found; the pages do not say whether you can mark something it missed.
Pricefive users, year one: about US$5,310
Per user, excluding VAT. Starter €49 a month billed annually (€59 monthly) covers document anonymization without the AI workspace, and is labelled "limited": Pro adds "unlimited pages", but Hexagone does not say how many pages Starter allows. Pro €79 a month billed annually (€99 monthly). Each licence is tied to one device. Enterprise on quote. 7-day free trial with unlimited use, no account or card.
Which plan we scored. Pro, because it is the cheapest plan with no stated page limit, the rule we apply to every tool. If Starter's limit is enough for your volume, five users cost about US$3,290 a year, and Hexagone's overall score would be 6.8, which would place it eighth.
Best for
French-speaking teams who want masking and an AI assistant in one app, and accept the online licence.
SourcesHexagone AIpricingtermssecurityFAQresearchlegal notice
Also considered
These run offline but missed the ten, on score or on our rules.
A desktop tool for freedom-of-information teams, with exemption codes, review circulation and an audit copy, at US$274 per user per year. Its detection is pattern-based, so names in running text depend on you.
Redacts patterns and word lists on Windows, offline, for a one-time US$79. It does not find names on its own (redaction help).
Removes what you mark, permanently, and searches preset patterns, but it has no automatic name detection and works on PDFs only (redaction guide). Fine for a few PDFs read line by line.
A local PDF editor with pattern-based redaction and no automatic name detection.
Runs on your own servers, with rule-based detection in nine European languages and OCR in 99. It is sold mainly to organizations, from €399 a month for five users.
A free French tool with a CamemBERT model and an encrypted local vault. It is French-only and its paid tier is not defined yet.
Offers on-premises deployment at enterprise scale, but no app an individual can install.
PII Anomalyzer
Listed as an offline tool at US$249 a year (LawNext), but we could not open a vendor website to check it.
Not ranked: cloud apps, developer libraries and cloud APIs
These tools fail the offline requirement or need developers, so they are not ranked here.
- Foxit PDF Editor+ finds names and IDs with Smart Redact, but Foxit says those files are processed in the cloud, on Azure.
- Redactable and iDox.ai are cloud web apps: your documents are uploaded to the vendor.
- VIDIZMO Redactor can be deployed on premises, but it is a large evidence and video platform priced on quote.
- Presidio is a free, open-source library (MIT), now community-run after leaving Microsoft. It runs offline, but it is a toolkit for developers with no interface. In our published benchmark it caught 70.4% of sensitive words with its large models.
- Limina, formerly Private AI, is a strong detection engine that runs as a container on your servers, but it contacts the vendor for licence checks and needs developers (source).
- Azure AI Language, Google Cloud Sensitive Data Protection and Amazon Comprehend are cloud APIs that process data in the provider's cloud.
- Tonic Textual prepares large text sets for AI training, for engineering teams.
How to choose
Start with two questions: may the machine connect to the internet at all, and do you need the original values back after using AI?
You want to use ChatGPT, Claude or Gemini on confidential documents, get real names back, and manage a team
promptShield
Small budget, Windows, documents stay on the machine
Philter Desktop or CamoText
The machine must never connect to anything
CamoText or Redactwizard
Air-gapped hospital or public body, scans and handwriting
Redactwizard
You mask files on a Mac before using AI
Quieta
You need Linux
Stript or AiNonymizer
German or English legal work
Stript
You work in Polish
promptShield or Anonimizer
You redact video and audio as well
CaseGuard Studio (Ultimate plan)
You want an AI assistant in the same app
Hexagone AI
You redact a few PDFs a month by hand
Adobe Acrobat Pro (not ranked)
Whichever tool you choose, check the output. Every tool here misses something sometimes, and a person reviewing the result before the file leaves is the cheapest safeguard.
FAQ
What is the best offline anonymization software in 2026?
It depends on what you redact. For audio, video and mixed media, CaseGuard Studio is the most complete offline suite. For text, including text in scans and images, promptShield scores highest at 8.2 out of 10, on the rules set out above. Redactwizard, built for networks with no internet at all, follows at 7.7, level with Philter Desktop, Quieta and CamoText, the best low-cost choices.
Does "offline" mean the app never connects to the internet?
Not always. Most of these tools process documents locally but still check their licence online. CamoText states it makes no network connections. Philter's only call is a manual update check. Redactwizard is designed for air-gapped use, and CaseGuard says it supports air-gapped networks. promptShield, Stript, AiNonymizer and Hexagone AI check their licence online on a schedule. Each review says how often.
What is the difference between redaction and anonymization?
Redaction removes data, usually under a black box. Anonymization is the broader goal of making a person unidentifiable. It can mean removing data, or replacing it with codes, labels or realistic fake values. Replacement keeps the document readable, which matters when an AI tool or a colleague has to work with it.
Is Adobe Acrobat enough to anonymize documents?
For a few PDFs that someone reads line by line, yes. Acrobat removes what you mark, permanently. But it does not find names or addresses by itself, so on long or numerous documents the risk is what nobody marked.
What is the best free anonymization software?
promptShield has a free browser version that processes files locally. Philter Desktop is free for personal use, Quieta has a free plan of 50 masks a month, and Anonimizer allows 5 documents a month. For developers, Presidio is free and open source.
Can these tools anonymize a database or spreadsheet for statistics?
That is a different job: you protect a whole dataset, not individual words. Look at a statistical tool such as ARX, an open-source program for anonymizing sensitive personal data in datasets.
How often is this ranking updated?
Every quarter. Prices and features change; if you find a fact that is out of date, tell us and we will correct it.
Full score sheet and sources
Every sub-score is out of 10. Detection is scored on the documented scale above, not measured. Technical = detection × 25% + offline × 20% + workflow × 20% + formats × 20% + languages × 15%. Overall = technical × 60% + value × 40%, computed from the rounded technical score. Scores are rounded to one decimal, halves up. Ties are ranked by the technical score.
| Tool | Detection | Offline | Workflow | Formats | Languages | Technical | US$ / point | Value | Overall |
|---|---|---|---|---|---|---|---|---|---|
| promptShield | 9 | 9 | 10 | 6 | 9 | 8.6 | 174 | 7.5 | 8.2 |
| Redactwizard | 8 | 9 | 8 | 7 | 9 | 8.2 | 191 | 7.0 | 7.7 |
| Philter Desktop | 7 | 10 | 8 | 7 | 3 | 7.2 | 138 | 8.5 | 7.7 |
| Quieta | 7 | 8 | 8 | 7 | 5 | 7.1 | 118 | 8.5 | 7.7 |
| CamoText | 6 | 10 | 8 | 5 | 5 | 6.9 | 64 | 9.0 | 7.7 |
| Anonimizer | 5 | 7 | 10 | 8 | 3 | 6.7 | 152 | 7.5 | 7.0 |
| AiNonymizer | 4 | 6 | 7 | 7 | 3 | 5.5 | 41 | 9.0 | 6.9 |
| Stript | 7 | 7 | 8 | 9 | 3 | 7.0 | 376 | 6.0 | 6.6 |
| CaseGuard Studio | 8 | 8 | 9 | 10 | 5 | 8.2 | 2,773 | 4.0 | 6.5 |
| Hexagone AI | 9 | 6 | 8 | 7 | 6 | 7.4 | 718 | 5.0 | 6.4 |
| Objective Redactnot ranked | 3 | 8 | 8 | 6 | 3 | 5.6 | 245 | 7.0 | 6.2 |
| PDF-XChange Editor Plusnot ranked | 3 | 9 | 6 | 4 | 3 | 5.0 | 79 | 8.0 | 6.2 |
| Adobe Acrobat Pronot ranked | 3 | 8 | 7 | 5 | 4 | 5.4 | 222 | 7.0 | 6.0 |
Workflow for the three tools not ranked: all three let you mark redactions by hand. Objective Redact adds review circulation, batch processing and an audit copy (8); PDF-XChange lets you review marked redactions before applying them (6); Acrobat adds a search across a folder of PDFs (7). None restores redacted text.
How the facts were gathered. Each fact comes from the vendor's own website, documentation, terms or pricing page, and is linked in that tool's section. G2 and Capterra could not be read reliably, so no review-site ratings are quoted. The only detection number we measured ourselves is promptShield's, on our published benchmark: 48 documents in 8 languages, with the documents, labels, scorer and raw results public. Hexagone AI's and Stript's accuracy figures are the vendors' own.
Found an error?
If any fact here is wrong or out of date, about any tool, ours included, tell us. We will be glad to correct it, and vendors are welcome to send us the details they would like checked.
Found this useful? Let us know.