Feature comparison
Competitor data from Gretel's public documentation and NVIDIA NeMo Microservices documentation, 2026. Gretel.ai was acquired by NVIDIA in March 2025; its technology now ships as NeMo Data Designer and Safe Synthesizer inside NVIDIA AI Enterprise, and the standalone gretel.ai site now redirects to NVIDIA — verify before relying.
| Feature | anonym.plus | Gretel.ai (now NVIDIA NeMo) |
|---|---|---|
| Data leaves your device | Never — 100% on-device, by default | Yes, by default — Safe Synthesizer's column-classification step calls NVIDIA's cloud inference API (integrate.api.nvidia.com) unless you build and host your own LLM endpoint; Data Designer and Safe Synthesizer are cloud/GPU microservices, not a local app |
| Deployment | Single installable desktop app (Windows/macOS) | NVIDIA NeMo Microservices (Data Designer + Safe Synthesizer), delivered via NVIDIA AI Enterprise — typically run on Kubernetes with NVIDIA GPUs, in your cloud or a partner's |
| Offline / air-gap | Yes, by default — verified with no internet connection at all | Only if you self-host every inference endpoint on your own GPU cluster — possible in theory, but a significant infrastructure project rather than a setting you toggle on |
| Entity types | 340+ built-in | 55+ PII types documented for Safe Synthesizer, scoped to structured/tabular columns — it is not a general free-text document NER engine |
| Encryption | Local AES-256-GCM with an offline key vault | No public claim of local encryption or a key vault; the privacy guarantee is differential privacy applied to synthetic output, not encryption of your source documents |
| Account / login required | No account for core use — internet is needed only once, for initial license activation | Yes — access runs through an NVIDIA AI Enterprise / NGC account; the old self-serve Gretel.ai signup and free tier were retired after the acquisition |
| Pricing model | One-time license, no subscription | Sales-gated enterprise licensing through NVIDIA AI Enterprise, generally tied to GPU infrastructure; the legacy Gretel.ai freemium pricing page no longer resolves |
| Setup / DevOps | None — download, install, run. Built on Microsoft Presidio + spaCy under the hood | Requires NVIDIA GPU infrastructure, Kubernetes, and NeMo Microservices deployment expertise before generating a single synthetic record |
| Compliance angle | On-device processing removes the processor/transfer question entirely — no document data egress by design, verifiable yourself by running fully air-gapped | Differential privacy on synthetic output can reduce re-identification risk in training data, but your organization still owns the controller/processor questions for whatever GPU infrastructure and NVIDIA cloud services you stand up |
Gretel.ai strengths
- Differential privacy guarantees are a mathematically rigorous way to protect structured/tabular datasets used for AI training
- Purpose-built for generating synthetic data that preserves the statistical properties of real datasets — genuinely useful for training and fine-tuning ML models
- Now backed by NVIDIA's engineering resources and GPU-accelerated infrastructure since the March 2025 acquisition
- Deep integration with the NVIDIA NeMo generative-AI stack, a real advantage for teams already building on NVIDIA's platform
- Handles structured, time-series, and some unstructured text data at meaningful scale for AI and agentic-AI training pipelines
Gretel.ai limitations
- No longer available as an independent, self-serve product — folded into NVIDIA AI Enterprise with sales-gated pricing since the acquisition
- Built for generating synthetic training data, not for redacting a real document you already have — a fundamentally different job
- Requires NVIDIA GPU infrastructure and NeMo Microservices/Kubernetes expertise to deploy — a heavy lift for a small team or a single user
- Default column-classification step calls NVIDIA's cloud inference API unless you build and host your own LLM endpoint
- No desktop app, no simple installer, and no path to genuine offline/air-gapped use without significant custom infrastructure work
Why choose anonym.plus
- anonym.plus redacts the documents you already have — contracts, records, spreadsheets — it doesn't generate synthetic replacements for them; different job, same goal of protecting people's data
- 100% on-device processing — no document content ever leaves your machine, verifiable by disconnecting the network and running it anyway
- 340+ PII entity types detected out of the box, no data-science pipeline or GPU cluster required
- Local AES-256-GCM encryption with an offline key vault — keys never touch a server
- Zero outbound network calls during document processing — no document data egress by design, verifiable yourself by disconnecting the network entirely
- One-time license, no subscription and no sales process — internet is needed only once, for initial activation
- Built on Microsoft Presidio + spaCy under the hood, packaged as a ready-to-run desktop app — no NVIDIA account, no Kubernetes, no infrastructure to stand up