DataCebo ships SDV 2.0 so companies can own a generative model of their databases

DataCebo released SDV 2.0, software that lets an organization build a generative relational model of its own databases inside infrastructure it controls — capturing structure, relationships, and business rules across tables — then use that model for synthetic data, scenario generation, and AI-agent training without moving production rows. Self-service pricing starts at $500/month; available now.

SOFTWARE desk — same-day primary product release for private generative models of enterprise databases, a concrete path to train agents and test systems without shipping production data.

What the company means by generative relational models: models trained from scratch that learn a database as a connected whole — statistical patterns, structure, relationships, context, and business rules spanning tables. The company pitch: large language models (LLMs) bring the world’s intelligence to the enterprise; generative relational models bring the enterprise’s intelligence to AI. File that definition and that pitch as DataCebo’s. This desk did not train a model.

How SDV 2.0 is meant to work, still company: teams can build that model from a representative subset of their data, typically in minutes to an hour, inside infrastructure the organization controls. When they connect the software to a database or a collection of files, DataCebo says SDV 2.0 can detect schemas and relationships — including primary, foreign, and composite keys — identify formats and context in the values, detect and enforce business constraints, tune the generative model, and generate task-specific datasets including rare events and edge cases. A schema is the map of tables, columns, and how they join. A primary key is the unique ID for a row. A foreign key is a pointer from one table to another. A composite key uses more than one column as that ID. File that automation list as DataCebo’s. This desk did not connect a warehouse.

Connects directly to Oracle, SQL Server, BigQuery, Spanner, and AlloyDB, the company said, and runs within the customer’s environment. File those connectors and the in-environment line as DataCebo’s. This desk did not sit on a customer network.

Availability, company: SDV 2.0 is available now. Self-service, consumption-based pricing begins at $500 per month for unlimited tables. Signup is at portal.datacebo.com/signup. File the now / $500 / unlimited-tables terms as DataCebo’s. This desk did not open an account or audit a bill.

Company-attributed customer examples — not independent Bad Signal audits: ING Belgium generated 10,000 synthetic payments in two minutes, which the company says is 100 times the test coverage in less than one-tenth the time previously required. Epiconcept created a synthetic database in 55 minutes and used it to identify optimizations that improved query performance 105 times. File those names and figures as DataCebo’s. This desk did not watch either run.

Roots, still company: the technology originated at MIT’s Data to AI Lab. CEO Kalyan Veeramachaneni is a Principal Research Scientist at MIT. CPO is Neha Patki. They led the team that created the Synthetic Data Vault, the source-available platform for generating tabular synthetic data. The company says the open SDV ecosystem has more than 18 million downloads, more than 5,000 paper citations, and more than 30,000 data scientists. File the lab origin, those titles, and those counts as DataCebo’s. This desk is not treating MIT as an institutional endorser of the product beyond the founder affiliation and lab origin stated in the release.

About-box, company: DataCebo is backed by Link Ventures, Uncorrelated Ventures, and Zetta Venture Partners. File those names as DataCebo’s. This desk did not see a cap table. The release does not print a company valuation or a recurring-revenue figure. This desk is not inventing either.

Plain English for the rest of the card: synthetic data = fake-but-realistic records that keep the shape of real tables so engineers can test and train without using real customer rows. generative relational model = one model that understands how all the tables fit together, not one narrow slice. SDV = Synthetic Data Vault. schema = the map of tables, columns, and how they join. primary key = the unique ID for a row. foreign key = a pointer from one table to another.

PRIMARY here: DataCebo’s 15 Sep 2026 PR Newswire company release — Tier A PRIMARY company source, the original record. The company site and signup portal are product-home / signup context, not a second originating newsroom. The SDV 2.0 / SDV Enterprise release, the generative-relational-model definition, the in-environment / minutes-to-an-hour training picture, the schema / key / format / constraint / tuning / rare-event automation list, the Oracle / SQL Server / BigQuery / Spanner / AlloyDB connectors, the available-now / $500-per-month / unlimited-tables terms, the ING Belgium and Epiconcept company-attributed examples, the MIT Data to AI Lab origin, the Veeramachaneni / Patki titles, the more-than-18-million-downloads / 5,000-citations / 30,000-data-scientists ecosystem lines, and the Link / Uncorrelated / Zetta backers are company-attributed. Customer case figures stay company-attributed — not independently verified here. MIT affiliation is founder / lab origin, not an institutional product endorsement. NOT claimed: a valuation, ARR, regulator certification, independently verified customer-speed figures, that this desk tested SDV 2.0, a stock tip, or investment advice. Distinct from the already-filed testmu-kane-assurance, cyberhill-cerebro-llms, cockroach-continuum, keewano-12m-keewanodb, wso2-agent-manager-ga, g5-labs-14m-seed, lectra-apogy, and gemini-3-8-live.