NVIDIA NeMo Data Designer

license

Apache-2.0(OSI, NVIDIA-NeMo/DataDesigner LICENSE)

source

public(pip install data-designer, the published framework runs locally against any configured inference provider)

core-gated

ungated(the NeMo Platform adds infrastructure around the published library rather than withholding generation from it)

NVIDIA's own documentation settles this. The NeMo microservices Data Designer page states that "The service is built on the open-source NVIDIA NeMo Data Designer library" and links to github.com/NVIDIA-NeMo/DataDesigner, which is Apache-2.0 and installs with `pip install data-designer`. The repository ships data-designer-config, data-designer-engine and data-designer, so the generation engine itself is published. What the NeMo Platform adds on top is infrastructure rather than function: inference routed through the NMP gateway, distributed job execution with monitoring, NMP fileset seeds, NMP artifact storage and the NMP Secrets service. None of that is functionality withheld from the published source, which is the test that decides this score. So the core is ungated - a managed service sold alongside a complete open core - and the product scores 5, open source.

• https://docs.nvidia.com/nemo/microservices/latest/data-designer/index.html recorded 2026-08-11

  NVIDIA's own NeMo microservices documentation for Data Designer states that "The service is built on the open-source NVIDIA NeMo Data Designer library" and links to github.com/NVIDIA-NeMo/DataDesigner. The library is installed locally to build configurations; the microservice is one way to execute them.

• https://raw.githubusercontent.com/NVIDIA-NeMo/DataDesigner/main/LICENSE recorded 2026-08-11

  Apache License Version 2.0, January 2004

• https://raw.githubusercontent.com/NVIDIA-NeMo/DataDesigner/main/README.md recorded 2026-08-11

  README titled "NeMo Data Designer" with an Apache-2.0 badge. Quick Start is "pip install data-designer". Describes the full generation framework -- statistical samplers, dependency-aware column generation, Python/SQL/custom validators, LLM-as-a-judge scoring, preview mode -- and refers to "slow self-hosted endpoints", so the published package runs against any configured inference provider rather than requiring an NVIDIA hosted service.

• https://techcrunch.com/2025/03/19/nvidia-reportedly-acquires-synthetic-data-startup-gretel/ recorded 2026-07-21

  NVIDIA acquired Gretel (Mar 2025); tech now shipped as NeMo Data Designer / NeMo microservices. Third-party summary, retained only for the acquisition history, and not relied on for any recorded key.

• https://github.com/NVIDIA-NeMo/DataDesigner recorded 2026-08-12

  Apache-2.0, 2.1k stars, packages/ ships data-designer-config, data-designer-engine and data-designer; README documents pip install data-designer and local generation against NVIDIA Build, OpenAI or OpenRouter providers

416,634 downloads in the trailing 30 days on the declared PyPI artifact `data-designer`, which bands at 100K-1M, level 3. The package is unambiguously the product's own rather than a same-named stranger: the quick start in NVIDIA-NeMo/DataDesigner's README is `pip install data-designer`, and for a Python library PyPI is the primary distribution channel rather than a minority one.

• https://pypistats.org/api/packages/data-designer/recent recorded 2026-08-13

  last_month 416,634 (last_week 145,088; last_day 24,257)

• https://raw.githubusercontent.com/NVIDIA-NeMo/DataDesigner/main/README.md recorded 2026-08-13

  quick start reads `pip install data-designer`, tying the PyPI package to this product

A capable synthetic-data engine, scored on its current NVIDIA-productized form. The transformation and augmentation workflows and the differential-privacy heritage from Gretel no longer appear on the NVIDIA page, so they do not count toward the score.

• https://www.nvidia.com/en-us/ai-data-science/products/nemo/ recorded 2026-08-13

  NeMo Data Designer is described as creating domain-specific synthetic datasets from scratch for building and evaluating specialized agents