anonymize.solutions vs Gretel
Gretel was a synthetic-data company: it generated new records that behave like the originals, for training and testing models. In March 2025 NVIDIA was reported to have acquired it, and gretel.ai now forwards to NVIDIA’s synthetic-data page. anonym.legal and anonym.plus work on the real text instead — the document or prompt you actually need an answer about.
Gretel today, from NVIDIA and the press
Quoted from NVIDIA’s page that gretel.ai now forwards to, and from the press, read on 24 September 2026. Anything those pages do not say, we do not claim.
| What NVIDIA and the press say | |
|---|---|
| Status | gretel.ai forwards to NVIDIA’s page “Synthetic Data Generation for Agentic AI”, which does not mention Gretel by name |
| What NVIDIA offers there | NeMo Data Designer and NeMo Safe Synthesizer, which “creates privacy-safe versions of sensitive data with default configurations designed to meet data privacy regulations such as HIPAA and GDPR” |
| The acquisition | reported in March 2025; terms were not disclosed by the parties |
Sources: nvidia.com, synthetic data generation ↗ · TechCrunch, 19 March 2025 ↗
Different jobs
| Gretel | anonym.legal / anonym.plus | |
|---|---|---|
| Built for | Generating new, synthetic datasets for training and testing | Anonymizing a real text so it can be used |
| What you get back | Records that resemble the originals | Your text, with the personal data replaced — reversible if you choose |
| Typical user | Machine-learning and data teams | Anyone who sends documents or prompts to an AI tool |
| Where it runs | NVIDIA’s NeMo stack | anonym.legal on our servers in Germany; anonym.plus on your machines |
Where each is the better answer
Gretel fits better when
- You need large training or test datasets that no real person is in
- Your team already builds on NVIDIA’s NeMo stack
- The goal is a model, not an answer about a specific document
anonym.legal or anonym.plus fits better when
- You need an answer about this contract, this record, this ticket
- The authorised reader has to get the real names back afterwards
- The text must never leave the machine (anonym.plus)
Using both
Synthetic data replaces the dataset; anonymization keeps the text and removes who is in it. A team can use synthetic data to train and anonymization to work with live documents — the two do not overlap much.
Not sure which layer your problem is on?
Tell us where the text is allowed to go and what it is. We will say whether it is ours to solve.