Hidden Tokens vs Presidio
Presidio and Hidden Tokens both find personal data and replace it. They are built for different people: Presidio for developers building a pipeline, Hidden Tokens for professionals who want to use Claude or ChatGPT safely today.
- Presidio is an open-source SDK, started at Microsoft and now transitioning to a community-owned project, for detecting and anonymising PII in text, images and structured data.
- It is a building block: you write code, choose models and operators, and deploy it yourself (Python, Docker, Kubernetes, HTTP service).
- Hidden Tokens is a finished desktop product for Mac and Windows: watched folders, Claude Desktop connector, browser extension, consistent codes per case, automatic restoration in Word and Excel.
- On our synthetic test set of French and Moroccan documents, Presidio's default configuration detected 77.5% of direct identifiers; Hidden Tokens 2.28.0 detected 100.0%. Method and per-pack scores.
- Choose Presidio to build your own pipeline; choose Hidden Tokens to give non-developers a working, provable workflow.
What Presidio is
According to its documentation (checked on 24 October 2026), Presidio provides identification and anonymisation modules for private entities in text and images: an analyzer for text, an anonymizer with several operators, an image redactor using OCR, and a module for structured and semi-structured data. It can run in Python or PySpark, in Docker or Kubernetes, or as an HTTP service. The project was started at Microsoft and is transitioning to community ownership. Its documentation warns that automated detection gives no guarantee of finding all sensitive information, which is true of every tool in this category, Hidden Tokens included.
Side by side
| Presidio | Hidden Tokens | |
|---|---|---|
| Type | Open-source SDK and services | Desktop app (Mac, Windows) with connector and browser extension |
| Who sets it up | Developers | Anyone: install, pick a folder |
| Where it runs | Wherever you deploy it | On each user's computer; optional shared registry on your file share |
| Word, Excel, PDF handling | Text, images and structured data; file handling is yours to build | Built in, including restoring values in Word and Excel output |
| Consistent codes across a case file | To build | Built in, per case and across the organisation |
| Use with Claude, ChatGPT, Gemini, Copilot | To integrate | Claude Desktop connector and browser extension |
| Sector rules | Custom recognisers you write | Six industry packs, plus custom rules |
| Direct identifiers detected, HT-Bench test set | 77.5% (presidio-analyzer 2.2.358, default recognisers) | 100.0% (connector 2.28.0, industry pack on) |
| Proof and admin | To build | Local AI log, signed proof pages, coverage report, signed admin policy |
| Cost | Free software, engineering time | €190 to €390 per seat per year |
Detection on French and Moroccan documents
We ran both tools on the held-out test split of HT-Bench, our synthetic benchmark: 366 documents from six industries plus general documents, 2,523 direct identifiers (names, Moroccan CIN, ICE and RIB, French NIR and SIRET, phone numbers, IBANs, cards…) and 3,029 decoys that should stay readable, scored by the same code on 25 October 2026.
| Presidio | Hidden Tokens | |
|---|---|---|
| Direct identifiers detected | 77.5% (1,956/2,523) | 100.0% (2,523/2,523) |
| Precision (coded passages that cover a labelled value) | 54.7% | 87.4% |
| Decoys coded by mistake | 37.6% | 5.8% |
Presidio configuration: presidio-analyzer 2.2.358 as shipped, AnalyzerEngine with its default recognisers, spaCy 3.8.7 with en_core_web_lg and fr_core_news_lg 3.8.0, each document analysed in English and in French, union of the spans, no score threshold, offline. Its defaults include no recogniser for Moroccan formats; a team can add its own. This compares Presidio's defaults with a finished product, on synthetic documents we wrote, so read it as our measurement, not an independent audit, and not as a score on real files. Per-pack scores, method, limits and the raw JSON are on the benchmark page.
Which one fits
- Choose Presidio if you are building an AI product or data pipeline, have engineers to integrate and maintain it, and need full control of the detection logic.
- Choose Hidden Tokens if lawyers, accountants, HR or support staff need to use AI on real files now, with consistent codes, automatic restoration and proof, without a development project.
Frequently asked questions
Is Presidio free?
Presidio is open source and free to use. The cost is engineering time: integration, tuning recognisers for your documents, deployment and maintenance.
Can Presidio put the real values back?
Presidio provides anonymisation operators, including encryption, and documents a matching decryption step for reversible cases. Building a full round trip (consistent codes across a case file, restoring values inside Word or Excel output) is up to you.
Does Hidden Tokens use Presidio?
No. Hidden Tokens runs its own engine, built for documents such as case files, invoices and payroll exports, with industry packs and per-column handling of numbers.