MOSTLY AI Synthetic Data SDK

Manufacturer / project: MOSTLY AI SDK project; brand assets: Syntho

When does it fit?

Consider the SDK when representative training data is available and you can evaluate distributions, relationships and downstream utility. Local experimentation is documented. 1 · 2

Fit still depends on your mandatory requirements.

What should you check before choosing?

Learned data may miss rare business exceptions or exact prescribed transactions. Check privacy risk rather than assuming synthetic output is anonymous. Brand transfer does not establish support or continuity of every former hosted service. 1 · 2 · 3 · 4 · 5

What must your PoC prove?

Hold out source rows before training. Measure rare-class coverage, relationships and leakage risk. Export and import the generator into a clean pinned environment; test random_state under the actual CPU/GPU setup.

Acceptance plan, not a product test result. Execution scope and limits.

Deployment and license

Python LOCAL mode on your infrastructure, or CLIENT mode with a remote endpoint. Optional connectors and models need additional dependencies. 1 · 2 · 3 · 4 · 5

The public SDK is Apache-2.0. Commercial platform services and support terms are separate; confirm the current supplier and entitlement. 1 · 2 · 3 · 4 · 5

Confirm the entitled version and operating requirements.

How you work with it

Train a generator, sample datasets and evaluate their utility. Export/import can move generators between SDK environments. This does not prove compatibility with another generation engine. 1 · 2 · 3 · 4 · 5

Export does not establish compatibility with another runtime.

Documented capabilities

Documentation is not a measured benchmark. Unclear is missing evidence; the comparison matrix explains the gap.

Sources and scope

Sources can cover different releases; they do not establish measured performance.

Use the same acceptance checks for your shortlist