Compile your intent once with an LLM. Then decide on every record forever — classification, scoring, routing — as type-safe JSON with probabilities and confidence, on your own hardware, at zero marginal cost.
The LLM runs once, at compile time. Production is a frozen, self-contained Python artifact: a small local decision model arbitrates every record, nothing is generated, every output is bounded, and low-confidence records abstain instead of guessing.
DEVELOPERS.md — architecture, compile lifecycle, calibration semantics and caveats, deployment, model swapping, full API reference.
