MOSTLY AI
Synthetic data platform with an Apache-licensed SDK that trains on your own machine
Quick Overview
| Company | MOSTLY AI (brand owned by Syntho B.V.; platform operating entity not verified) |
|---|---|
| Category | Synthetic Data Platforms |
| Headquarters | Vienna / Amsterdam, Austria |
| Founded | Not stated |
| EU Presence | Yes - Austria and Netherlands (EU) |
| Data Location | Your own compute with the local SDK; hosted version at app.mostly.ai |
| Open Source | Apache 2.0 (SDK, per vendor) |
| GDPR Compliant | Yes |
| Self-Hosting | Yes, local SDK and Kubernetes or OpenShift platform |
| Main Features | TabularARGN model, differential privacy option, multi-table, time series, text data, QA reports, rebalancing, AI assistant |
| Pricing | Free hosted version at app.mostly.ai; enterprise pricing on request |
| Best For | Data scientists who want to start in Python and move to a shared platform later |
| Replaces | Gretel, NVIDIA NeMo Data Designer |
Detailed Review
MOSTLY AI is best known for its Synthetic Data SDK, which installs with pip and trains a generator on your own hardware. The home page describes it as a permissive open-source project under Apache v2.
On 9 June 2026 Syntho announced it had bought the MOSTLY AI brand, and the site now reads MOSTLY AI powered by Syntho.
Local first, platform second
In local mode the SDK trains and generates on your CPU or GPU, and Docker and air-gapped installs are covered in the docs. A client mode connects to a remote platform, and locally trained generators can be uploaded for sharing.
What the SDK can model
Documentation lists mixed data types including categorical, numerical, geospatial and text, plus single-table, multi-table and time-series data. Differential privacy is offered as an advanced training option, alongside rebalancing and imputation.
The platform side
The commercial platform groups four data types: real-world data analysed with an AI assistant, mock data, synthetic data and simulated scenarios. It is described as deployable on Kubernetes or OpenShift.
Limitations to Consider
The legal entity behind the platform contract could not be confirmed from pages that were readable, and the brand transfer to Syntho in June 2026 means terms may change; check the order form. No prices are published for the enterprise platform. The SDK requires Python skills, unlike the interface-led Syntho engine or Dedomena.
Who Should Use MOSTLY AI
Python-literate data science teams who want an open-source generator they can audit and run offline, with a managed platform available if collaboration becomes necessary.
Pros and Cons
Pros
- Apache-licensed SDK
- Trains on your own CPU or GPU
- Handles multi-table, time series and text
- Differential privacy option
- Free hosted version
Cons
- Contracting entity not confirmed
- Ownership recently changed
- Enterprise prices not public
- Needs Python knowledge for the SDK
Alternatives to MOSTLY AI
Other European synthetic data platforms worth comparing before you decide:
Octopize
France · Avatar anonymisation with synthetic data and public price tiers
Syntho
Netherlands · Masking, rule-based and AI-generated data in one engine
Dedomena
Spain · Madrid platform pairing validation, anonymisation and synthetic data
Aindo
Italy · Trieste synthetic data for clinical evidence and public health
See all European synthetic data platforms or compare them directly against Gretel on our alternatives page.
Frequently Asked Questions
The Synthetic Data SDK is described on the vendor's home page as a fully permissive open-source project under an Apache v2 licence.
Syntho announced on 9 June 2026 that it acquired the MOSTLY AI brand; the site is titled MOSTLY AI powered by Syntho.
Yes. Local mode trains and generates on your own compute, and the docs cover Docker and air-gapped installation.