An alternative to a legal assistant: ndaact.ai compared
Keeping consumer ChatGPT or Claude and redacting locally costs 8.50 EUR excl. VAT per month, 2 devices included, while Doctrine and anonym.legal do not show a price on the pages we read on 11 October 2026.
This page compares four approaches for a law firm, a notary office or a medical practice that wants to use AI on real files: ndaact.ai, Azure OpenAI, a legal assistant (Doctrine) and a data masking tool (anonym.legal). Every claim about a third party comes from the official page cited, read on the date shown. When a page says nothing, the cell says so too.
| Criterion | ndaact.ai | Azure OpenAI (Microsoft) | Legal assistant (Doctrine) | Masking tool (anonym.legal) |
|---|---|---|---|---|
| Where the data runs | Detection and replacement run on your machine. The text, with markers in place of the detected entities, then goes to the provider of the model you use. | Processing in the geography you choose, except Global or DataZone deployments where it can happen elsewhere. A sample may be kept for abuse detection unless a modified-monitoring exemption is approved. | Not stated on the pricing page we read. | Cloud version: Hetzner data centres in Germany. Desktop app: processing on the device. |
| Reversibility | Markers such as {{PERSON_a1b2}} are swapped back to the original values in the reply, using an encrypted mapping table stored on your machine. |
Not described in the documentation we read. | Not described on the page we read. | The site states that tokenisation can restore the original values. |
| Languages | Built first for French (NIR, SIRET, IBAN, addresses). Other languages are described in the documentation. | Not recorded on the page we read. | Not recorded on the page we read. | 48 languages, according to its site. |
| Monthly price excl. VAT | 8.50 EUR per month or 75 EUR per year, 2 devices, free trial. | Pay-as-you-go per token processed, or provisioned throughput. No per-seat price. | No price shown. "Avocat individuel" plan with a 7-day trial; firm and in-house legal plans on demo request. | No price shown on the home page we read. |
| Model used | The one you already use: ChatGPT or Claude on the web. ndaact.ai is not an assistant and only ships a local detection model. | The models offered by Azure, including OpenAI models, hosted by Microsoft. | Not stated on the page we read. | A document processing tool, not a conversational assistant, according to its site. |
| Vendor lock-in | You keep your AI tool. Subscription can be cancelled at any time. | Resources are created in your Azure subscription. | Individual subscription. Add-on modules need a separate subscription. | Not recorded. |
What the table does not say
No single building block sets these tools apart. A masking tool can already run on the device and restore values, a cloud service can commit to a processing region, a legal assistant can cite case law. The value of ndaact.ai lies in the combination: local detection, reversible replacement with markers, settings built for French, and your usual AI tool kept in place.
ndaact.ai is an aid, not a guarantee. Detection can miss entities, and an indirect identifier left in free text can still point to someone. The data remains personal data under the GDPR, because the operation is reversible for you. Proofread the text before sending when the stakes are high.
Limits of this review
ChatGPT Enterprise and Gino are not in the table: their official pages could not be read on 11 October 2026. They will be added at the next review, after reading them directly. Prices and terms change: check them with the vendor before any decision.
Sources read on 11 October 2026
- Microsoft Learn, data, privacy and security for models sold by Azure (page updated on 5 June 2026).
- Microsoft Azure, Azure OpenAI pricing.
- Doctrine, pricing.
- anonym.legal, home page.
- ndaact.ai, home page for the price and how it works.
Source: Microsoft Learn, 5 June 2026.
This page is a translation. The French version prevails.