LiteLLM Proxy - quick start documentation

Fetched 2026-10-04. The proxy is described as an 'LLM Gateway' managing three things verbatim: 'Unified Interface: Calling 100+ LLMs ... in the OpenAI ChatCompletions & Completions format'; 'Cost tracking: Authentication, Spend Tracking & Budgets Virtual Keys'; 'Load Balancing: between Multiple Models + Deployments of the same model - LiteLLM proxy can handle 1.5k+ requests/second during load tests.' The 1.5k RPS figure is the project's own load-test claim, i.e. a vendor number. Config shape is model_list entries with model_name (user-facing alias) plus litellm_params (model, api_base, api_key). Also documents vLLM as a first-class target (model: vllm/..., api_base: http://your-vllm-base:3000/v1, api_key 'none'). Logging knobs: --debug, --detailed_debug, LITELLM_LOG env var. Install floor: 'LiteLLM 1.84.0 and newer require Python 3.10 or higher', and a bare pip install on an older interpreter 'silently resolves down to the last release whose requires-python still allowed it (1.83.9 ...) with no error' - an installation-time silent downgrade worth recording separately.