Your database stays yours. The schema is the configuration.
Platform pitches usually mean a proprietary data store, a proprietary language, and a hostage negotiation in year five. This one runs over your SQL Server, generates nothing you can't read, and is built to be left.
Start here.
Three pains, three mechanisms.
Low-code platforms trade six fast months for a decade of lock-in.
Forge is metadata over your own relational schema. The data model stays normal SQL Server. Application code is normal .NET and React. The scaffolding generates readable projects your team owns. Leaving would be unpleasant, like leaving anything, but your data and your code come with you.
Data access and the instant API →Every new service reimplements auth, audit, paging, and background jobs slightly differently.
One engine provides them uniformly: no-ORM async data path, auto-generated GraphQL, temporal history, typed events, scheduled connectors, tenant-scoped queue processors. Register a table and the platform's behavior is identical to every other table's.
Integration and automation →AI coding assistants generate plausible code that violates your platform's rules.
Scaffolded projects ship AI-readable convention files, and the MCP server gives assistants live schema introspection plus a structural audit. The assistant reads the real rules and the real data model instead of guessing at either.
The developer experience →Straight answers.
.NET 10 services over SQL Server with fully-async ADO.NET and no ORM. HotChocolate GraphQL, SignalR for real-time, Argon2 identity, Roslyn for sandboxed rule compilation, MCP for AI tooling. The admin client is Next.js and Fluent UI.
One chokepoint. Security, tenancy, and audit are enforced in a single data path that nothing can route around, and there's no ORM overhead or generated-entity drift between the app and the database.
SQL Server only, for now; the engine reads standard catalog metadata and PostgreSQL and Oracle are next on the same architecture. Collaborative draft editing is the newest major subsystem. If a limit matters to your evaluation, ask and you'll get the straight answer.
Validation rules are metadata: authored by prompt or by hand, compiled at runtime, executed in a sandbox, test-gated, approval-gated, and tenant-scoped. The core platform doesn't redeploy when a business rule changes.
Evaluate it like an engineer.
Bring a schema with an ugly corner. The assessment is more interesting when the data isn't a demo.