LiteLLM is an open-source AI gateway and LLM proxy. It offers a single OpenAI-compatible interface to more than a hundred model providers, so applications can switch between models without changing code. It is available as a Python SDK for use inside applications and as a proxy server that platform teams run centrally. The gateway adds virtual keys, spend tracking per user or team, rate limits, load balancing, caching, guardrails such as PII masking, and SSO to manage access across an organisation.
LiteLLM is developed by BerriAI, a company headquartered in San Francisco in the United States. The core gateway is released under the MIT licence, while enterprise features such as SSO and audit logs fall under a separate commercial licence. Organisations deploy it themselves with Docker, Helm or Terraform, including in air-gapped environments, and according to the vendor the self-hosted version sends no telemetry. The gateway therefore runs in your own infrastructure. Where prompts end up depends on the model providers you route to, so a European organisation can combine LiteLLM with EU-hosted or self-hosted models.