Project / Oct 01, 2026
AI Gateway Playground
A small lab for comparing model routing, fallback behavior, latency, and observability across LLM providers.
Status: building
This project turns AI Gateway concepts into observable experiments rather than a single polished demo.
Questions
- How should a gateway expose provider failures without hiding them behind retries?
- Which routing signals are useful beyond price and latency?
- How can traces connect a model request to an application-level evaluation?
Planned slices
- A unified request format for several model providers.
- Request-level latency, token, and error traces.
- Explicit fallback policies with visible decision logs.
- A small evaluation set for comparing behavior across models.