Gretel.ai vs Neosync

Quote on request

Free and open source

The old credit-based self-serve tier is gone. The successor microservices are licensed through NVIDIA AI Enterprise; NeMo Data Designer can be tried on build.nvidia.com.

The MIT-licensed code remains free to self-host. The paid cloud service no longer exists — neosync.dev is offline.

Everything, self-hosted

Emerging

NVIDIA (formerly Gretel) · since 2019

Emerging

Nucleus Cloud (acquired by Grow Therapy) · since 2023

• Formal differential-privacy guarantees rather than best-effort masking, now via DP-SGD in NeMo Safe Synthesizer.
• Synthetic data preserves statistical relationships useful for analytics and model training.
• Backed by NVIDIA's platform investment rather than a single startup's runway.

• Open source and self-hostable, so sensitive data never leaves your infrastructure.
• Preserves referential integrity across tables.
• The last release is feature-complete for common Postgres and MySQL anonymisation work.

• No longer buyable as Gretel — you are procuring NVIDIA AI Enterprise, which is far heavier than the old self-serve signup.
• Model training adds time and cost compared to rule-based masking.
• Safe Synthesizer trains on a real dataset, so it does not help with greenfield data that does not exist yet.

• Archived and unmaintained since August 2025 — no security patches, no new connectors, no support.
• The hosted cloud has been shut down, so self-hosting is the only option.
• Fewer database connectors than the incumbents.

Regulated enterprises already standardising on NVIDIA AI Enterprise. Teams that wanted Gretel's self-serve tier should look at Tonic or Neosync instead.

Realistically, nobody starting fresh. Existing self-hosted users can keep running it, but plan a migration to Tonic or a maintained alternative.