ε ∈ [0.1, 12]
Differential privacy
Add calibrated noise to gradients during training. Pick your ε per dataset; we'll report the budget consumed and warn before you exceed it.
Everything in Doppelset, on one page.
Pick a row. Open a feature. Or skim the lot — we hide nothing, we oversell nothing. If a checkmark isn't here, the feature isn't built yet.
Default-on
Privacy controls
60+ entities
Direct identifier vault
60+ entity types across 32 languages: names, emails, IBANs, ICD-10 codes, GPS coordinates, NIE numbers… stripped before model sees them.
k ≥ 30
Quasi-identifier protection
k-anonymity, l-diversity, and t-closeness checks run on every output. Records that look unique against background knowledge get re-sampled.
12 attacks
Memorisation test suite
Every generation runs a battery of attribute-disclosure and membership-inference attacks. The signed report includes pass/fail per attack.
Hybrid engine
Modelling
core
Tabular transformer
A masked-attention model that learns mixed categorical, ordinal and continuous columns jointly — no one-hot blow-up, no embedding hand-tuning.
default
Diffusion sampler
Iterative denoising at sample time gives sharper tails and rarer regimes than GAN-only approaches. Picked automatically for numerical-heavy tables.
multi-table
Relational coordinator
A second-pass model enforces foreign-key cardinality, parent/child structures, and join distributions across multi-table generations.
new
Time-series module
Captures seasonality, drift, change-points, and rare anomalies in a single state-space model. Generates irregular series too.
You can trust
Quality & evals
0–100
Statistical fidelity score
A single number, 0 → 100, that summarises 80+ tests: marginals, correlations, conditional distributions, downstream ML utility.
Distribution diff
Per-column overlap plots with KL/Wasserstein/JS distances. Inline annotations explain why a column drifted.
TSTR
Utility (TSTR) benchmark
Train-on-Synthetic, Test-on-Real benchmark across 14 model families. Pick a task — classification, regression, churn — get a number.
audit-ready
Signed run receipts
Every output ships with a JSON receipt cryptographically signed by Doppelset. Hand it to your auditor, paste it in your DPIA.
Pipe-shaped
Integration
py 3.9+
Python SDK
pip install doppelset · type-checked · async-first · streams sample() output so 4 B rows never live in RAM.
TypeScript SDK
@doppelset/sdk on npm. Works in Node, Bun, Deno, and Cloudflare Workers. ESM first.
OpenAPI
REST API
OpenAPI 3.1, OAuth 2.1, idempotency keys, retries, server-sent events for long-running jobs.
12 stores
Connectors
First-class connectors for Snowflake, Databricks, BigQuery, Redshift, Postgres, MySQL, S3, GCS, Airflow, dbt, Kafka, MongoDB.
Enterprise
Governance
Okta + 5
SSO + SCIM
Okta, Entra, Google, JumpCloud, OneLogin. Automatic provisioning, role propagation, just-in-time access.
Roles & projects
Granular RBAC. Synth-only seats for analysts, viewer seats for compliance officers, admin for platform owners.
SIEM
Audit log
Every generation, every config change, every download — append-only, hash-chained, exportable to your SIEM.
EU · US · VPC
Data residency
EU (Madrid, Frankfurt), US (Virginia, Oregon), or your own VPC. Data never leaves the region you pick.
Try it now
Ship faster. Stop arguing with legal.
Generate your first 100,000 synthetic rows in the next ten minutes. No credit card.