LiteLLM is two different products, and the free one is both

"LiteLLM" means two things that get conflated. There's the Python library — litellm.completion() as a uniform client over a hundred providers, living inside your app. And there's the proxy server — a standalone gateway your whole team points at, with virtual keys, spend tracking and routing config in one place.

Both are open source and free; the paid product is the managed/enterprise layer on the proxy. The choice between them is architectural, not financial: the library couples provider config to each codebase (fine for one app), the proxy centralises it (fine for many apps, or for rotating providers without redeploying anything).

The proxy is also the natural place for the hygiene that free-tier stacks need — fallback chains, rate-limit smoothing, and one config file where dead model IDs get fixed once instead of in every repo.

Setup for both modes and the fallback config that matters: https://toolfreebie.com/litellm-free-llm-gateway/