Gretel.ai vs Hoverfly

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.

Free and open source. You pay in engineering time and CI minutes.

Everything, self-hosted

Emerging

NVIDIA (formerly Gretel) · since 2019

Emerging

SpectoLabs · since 2016

• 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.

• Capture mode produces a realistic simulation from live traffic with no authoring.
• Latency injection makes it useful for performance and resilience testing too.
• Small footprint — a single Go binary.

• 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.

• Smaller community and slower development than WireMock.
• Proxy configuration in complex networks takes effort.
• Captured simulations need periodic refresh.

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

Teams needing lightweight service virtualisation against real recorded traffic.