LiteLLM
One OpenAI-shaped interface for 100+ model providers, plus a proxy that tracks spend per key.
LiteLLM normalises provider APIs behind a single SDK and ships a proxy server with budgets, rate limits and automatic fallbacks. It is the cheapest way to stop hard-coding one vendor into an application you will have to migrate later.
Best for: Multi-provider routing, spend control, fallbacks
Deploy: Self-hostable