LiteLLM vs OpenRouter
LiteLLM and OpenRouter solve the same problem (one endpoint, many providers, automatic fallback) from opposite directions: LiteLLM is open-source software you run, OpenRouter is a hosted service you pay per use. The right answer depends on whether you want to run a process.
LiteLLM: your proxy, your rules
LiteLLM runs wherever you put it: laptop, VPS, container. You bring provider keys, it exposes one OpenAI-compatible endpoint with retries, fallbacks, and load balancing. Cost: zero for the software, provider list price for tokens. Privacy: nothing leaves your infrastructure except calls to the providers themselves.
The historical friction is config authoring: model_list entries, api_base URLs, env var wiring. Our router builder removes exactly that step: order your providers, export LiteLLM YAML, run it.
OpenRouter: the hosted shortcut
OpenRouter is LiteLLM-as-a-service plus billing consolidation: instant, no process to run, uptime stats, one credit balance. The trade is structural: your prompts transit their servers and your cost scales per token forever.
For quick prototypes and for teams that refuse to run anything, that trade can be worth it. For anything sustained, self-hosting the routing layer is both cheaper and more private.
The 60-second LiteLLM setup
Add two or three provider keys in the Dracon key manager, open the router builder, stack them in priority order, pick LiteLLM YAML, export, and run `litellm --config litellm.config.yaml`. The exported config references env vars for keys, so secrets stay on your machine. Each provider in the chain has a setup guide if you need the key first.