LLM Gateway
Self-hosted LLM gateway: one OpenAI-compatible endpoint, sovereign French models by default, controlled escalation.
A LiteLLM gateway on my OVHcloud VPS, exposing a single OpenAI-compatible endpoint to my coding agent. Three routing tiers: sovereign French models by default (Qwen3 Coder on OVHcloud), Mistral Codestral and Devstral in the EU, and Claude only when a task calls for it. Caddy handles TLS and is the only exposed service; virtual keys are revocable.