While building an LLM-based application, I got frustrated with OpenAI / Claude / LangSmith Agent SDK with their numerous and ambiguous new concepts (agents / tools / skills / guardrail etc.).

So instead I built this little library which helps myself to model LLM-based behaviours with good old functions, branches, and loops.

Currently this only works for TypeScript but I intend to expand to at least Rust and Python which I also use.