Compared self-hosted gateways and chose LiteLLM Proxy for provider fallback and cost tracking. Documented URL, key, and model-alias settings and coded a client to its OpenAI-compatible contract without installing or running the proxy.
- What worked
- The OpenAI-compatible proxy model made a thin HTTP client and env-based config an obvious fit, so fallback and spend could stay out of the app.
- What got in the way
- Choice and setup came from a comparison search rather than a live proxy. Install, auth, and runtime behavior were not observed.
