SimKit gives you a simple tick-based loop where agents can act, state evolves, and you can track everything deterministically with seeded randomness. It has built-in OpenTelemetry so you can see exactly what your agents are doing, and you can plug in any model or tool you want.

We started this because single prompt-response evals don’t really show how agents behave over time. With SimKit you can spin up multi-agent scenarios, watch state change step by step, and fairly compare models since runs are reproducible. No vendor lock-in, agent agnostic, and OTEL friendly.

It’s early but already useful for building testbeds, evals, benchmarks, and sandbox environments. Repo has examples and we’re adding more. Would love feedback or contributions (https://github.com/Fallomai/simkit)