SC# stands for Sandbox C Sharp. It is a small, statically typed and interpreted sandbox language with C#-like syntax, created for Ai4Users to run compact LLM benchmark and validation scripts in an isolated, deterministic and host-controlled way.
SC# was created to solve one central problem: safely running untrusted scripts on a server without Docker containers, process sandboxes, virtual machines or other external isolation layers. Scripts are parsed, type-checked and executed in memory by the SC# runtime, without producing binaries, assemblies or intermediate files.
SC# is currently available only on Ai4Users. It is used here to run isolated, deterministic LLM benchmark and validation tests.
SC# is designed for small algorithmic tasks, validators, scoring rules, fake tool fixtures and LLM integration tests where the host application must remain in control.
Before a script runs, the runtime validates the strongly typed code and rejects syntax, type and entry-point errors. Accepted scripts are then executed by a tree-walking interpreter using a deliberately closed set of language features, built-ins and host-provided functions.
Ai4Users uses SC# to run tests written for LLM benchmark scenarios, validation checks and tool-use experiments. A test can inspect structured output, validate tool calls, compare values, keep limited virtual state and fail with a clear diagnostic while the platform keeps execution bounded.
This makes SC# a practical fixture language for benchmark code: enough structure for repeatable checks, with restrictions that keep scripts narrow, reviewable and safe to run.
The language includes only the features needed to express deterministic checks: typed values, control flow, collections, helper functions, assertions and controlled host-provided operations.
SC# is not C# and does not use the C# compiler, C# libraries or the C# runtime ecosystem; it only uses C#-like syntax where that makes compact scripts easier to read and write.