Gretel
Synthetic data platform for developers: generate privacy-safe tabular, text, and time-series data via API.
Acquired — still operating
NVIDIA reportedly acquired Mar 2025
About
Gretel is a developer-first synthetic data platform: its APIs and SDKs train generative models on your real data and produce artificial datasets that preserve statistical properties while protecting privacy through differential privacy and PII detection. It covers tabular, time-series, and unstructured text data, with quality and privacy scoring on every output, plus transform and redaction workflows. Use cases center on AI/ML training data, safe data sharing across teams, and regulated industries like finance and healthcare. In March 2025 NVIDIA reportedly acquired Gretel (nine-figure deal, above its $320M valuation) to strengthen its AI training data tooling — worth factoring into long-term dependency planning. Pricing is credit-based: a free developer tier (15 credits/month), then ~$2/credit pay-as-you-go, Team at $295/month, and custom Enterprise. Very large datasets can be slow to process and advanced tuning assumes ML familiarity.
What it can do
- Tabular, text, and time-series synthesis; differential privacy; PII detection and redaction; quality and privacy scoring; transform/augment workflows; APIs, SDKs, and cloud console; deployable via AWS/GCP/Azure marketplaces.
Setup
Sign up for the cloud console, connect data via the SDK or UI, train a synthetic model, then generate and validate output with built-in quality reports.
Prerequisites
Sample of representative real data; Python for SDK workflows (no-code console also available).
Use cases
Works with
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Synthetic data types (tabular, text, time-series), differential privacy, credit pricing (free 15 credits/mo, $2/credit, Team $295/mo) from pricing breakdowns.; NVIDIA acquisition (reported Mar 2025, nine figures, above $320M valuation) from press; flagged as reported, with dependency-planning caveat.; Limitations disclosed: large-dataset processing time, ML familiarity needed for advanced tuning..
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.