The
LiteLLM proxy exposes an OpenAI
/v1/responses endpoint and bridges to Chat-Completions providers automatically — so Responses-only clients like the Codex CLI can talk to Mistral, DeepSeek, Anthropic, Gemini and more. API keys are referenced as
os.environ/NAME and read by the proxy at runtime; they're never written into the YAML. Model names are best-effort for mid-2026 — verify with each provider.