From a spreadsheet
to a substation —
with receipts.
Generate realistic, safe data — for a spreadsheet or a whole factory — and get proof it’s correct.
One platform, from ordinary tables to live industrial systems. No real records are copied, and every run is sealed so you (or your auditor) can replay and verify it later — a range few tools offer.
| invoice_id | currency | grand_total | invoice_date | subtotal | tax | customer_id |
|---|---|---|---|---|---|---|
| 196 | USD | 0.00 | 2026-06-09 | 0.00 | 0.00 | 167 |
| 1288 | USD | 0.00 | 2026-06-11 | 0.00 | 0.00 | 432 |
| 1197 | USD | 40680.67 | 2026-07-02 | 37667.29 | 3013.38 | 167 |
| 926 | USD | 0.00 | 2026-06-07 | 0.00 | 0.00 | 918 |
| 1559 | USD | 2322.39 | 2026-06-11 | 2150.36 | 172.03 | 922 |
| 1741 | USD | 9167.04 | 2026-06-18 | 8488.00 | 679.04 | 171 |
| 380 | USD | 0.00 | 2026-06-30 | 0.00 | 0.00 | 930 |
| 1106 | USD | 0.00 | 2026-06-14 | 0.00 | 0.00 | 665 |
| 1925 | USD | 16229.88 | 2026-06-20 | 15027.67 | 1202.21 | 420 |
| 562 | USD | 16595.17 | 2026-06-02 | 15365.90 | 1229.27 | 167 |
Describe it. Get data. Verify every row.
You describe the data, the engine generates it under a sealed contract, and every generated row is checked against your rules before it ships.
Plain English, or your own schema.
Pick a mode — Mock, Synthesize, OT/SCADA, ICS Security — or let Auto choose. Type what you need, upload a CSV, or point at a connector. A cost estimate shows before anything expensive runs.
- Auto-routes to the right engine
- Cost shown before you spend
- Your data never leaves your tenant
What would you like to generate?
No identifier-shape values detected; dataset clean.
- Inspected rows
- 100
- Inspected attributes
- 15
- Repairs applied
- 0
- Scanner
- v1-row-level
Every row checked, then sealed.
The engine evaluates cross-field invariants on every generated row — arithmetic totals, declared enums, non-negativity — and records the result. Nothing ships silently: a failure is surfaced, a repair is logged. The whole run is hashed into a bundle you can replay and verify offline.
- Cross-field business rules, per row
- Reproducible from the same seed
- Offline verifier ships with the SDK
Water treatment pump station SCADA, 60 seconds, Modbus + OPC-UA, 8 signals: discharge pressure, suction pressure, motor temperature, VFD output, flow rate, tank level, and an over-pressure alarm. [Protocols: modbus, opcua] [Duration: 1m] [Poll: 1 Hz] [Streaming: off]
small_wwtpRun a SCADA simulation for the water treatment pump station.
Verify the quality and provenance of the simulated data.
A live substation, on real OT protocols.
This is a real run: a water-treatment pump station simulated on Modbus and OPC-UA, at the same wire level your equipment sees. The output is physics-checked (not canned CSVs), packet captures included, and sealed into an integrity trail — the same evidence chain a tabular job produces. Industrial OT and ICS attack data as a first-class product is rare in synthetic data. It’s core to us.
- 6 OT protocols at binary-spec level
- Physics-honest process models, not fixed CSVs
- ICS attack datasets with per-event ground truth
Most tools hand you a quality score. We hand you a receipt you can replay.
Because generation is deterministic, the same request and seed produce byte-identical output — so an auditor can re-run it a year from now, on a different cluster, and get the exact same hash. Stochastic ML synthesizers can’t make that promise. You get the data, the per-row checks, and the cryptographic proof it all lines up.
- Same seed → identical hash
- Sealed contract roots the chain
- Offline verifier ships with the SDK
The same job, held to a higher standard.
No competitor is named below — this is the honest category picture from our own research. Where most synthetic-data tools stop at tabular data and a score, we go further.
An agent that plans, runs, and self-heals — and asks before it spends.
Hand it a goal in plain English. It breaks the work into steps across every engine, runs them, repairs what fails, and stops for your approval before any expensive step. Every decision, tool call and cost lands in the same sealed evidence trail. The AI Assistant is the same brain in a chat box, for people who don’t want the SDK.
Everything you can make.
Ten surfaces, one sealed-job contract, one evidence chain, one tenant model. A SCADA run can feed a synthesis job; the autonomous agent can drive any of them.
Mock Data
Prompt → a realistic, sealed dataset in under a minute.
Synthesize
Learn from your own data; higher-fidelity, quality-gated output.
Autonomous Data Scientist
Plan → run → self-heal across engines, with a cost gate.
AI Assistant
Drive the whole platform in plain English, with visible cost.
Healthcare FHIR
HL7 FHIR R4 bundles, shipped vocabularies, zero PHI.
Virtual SCADA
Six OT protocols at wire-level; physics-honest process models.
Virtual PLC
Air-gapped controller image you can run inside your VPC.
ICS Security
MITRE ATT&CK ICS attack data with per-event ground truth.
Physics Model
Continuous-time paths with conservation laws enforced.
API · SDK · CLI
One typed surface: REST, Python SDK, the rady CLI, connectors.
Real records that can't leave the building.
Different teams, one problem: you need realistic data, and you need to prove what you shipped.
SMEs & startups
Stand up realistic test data in minutes — without touching real customer records.
Enterprise data teams
Reproducible training/test data with a full audit trail every regulator will accept.
Security & OT teams
SCADA, PLC and labelled ICS attack data your SOC can score against — repeatably.
Developers
Wire it into your pipeline through a typed API, SDK, CLI and 14 connectors.
Security by architecture.
Isolation, encryption, and a cryptographic evidence chain are built into the architecture — not settings you switch on.
Per-tenant keys, prefixes and row-level filtering — no path between tenants.
Engineered to ISO 27001 / SOC 2 control expectations, GDPR & UK GDPR.
Data protected in transit and at rest; connector secrets never sit in config.
A public status page backed by real probes.
Bring a real dataset. Keep the receipt.
Generate against a representative dataset and one open question, end to end. You keep the sealed evidence bundle, the quality report, and a sandbox key to keep iterating.