What it is
One SDK for many decision models, plus a playful way to compare them.
What it does: Lets you swap decision providers behind one call and watch models compete.
Why it belongs in the showcase: Provider-neutral typing with budgets and cost tracking built in.
Caveats: Multiple keys; game scores are not task scores.
Evidence
Public MIT repo (2★ at intake) created 2026-10-17. README lists five providers marked live-tested, typed answers that may be refusals, usage with cost (provider-reported or from list prices), and an arcade server that stores every call in SQLite, enforces daily and per-visitor budgets, and ranks models in the Arena and Attention games.
Caveats
- Needs keys for each provider you want to compare
- Cost for some providers is estimated from list prices
- Games are benchmarks for fun, not task accuracy
Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.