hwdsl2/docker-litellm

This project gives you a single, private gateway for interacting with over 100 different AI models. Instead of juggling separate accounts, API keys, and software tools for OpenAI, Anthropic, Google Gemini, and others, you connect your apps to one place. It makes all these different providers look and act like OpenAI, so any software already built for OpenAI can seamlessly use any model you choose. Under the hood, it runs as a self-hosted proxy. When you first start it, the software automatically generates a secure master key and sets up a configuration file. You simply provide the API keys for the AI services you want to use, and the proxy handles routing your requests to the right provider. It also includes tools to create "virtual keys," which are temporary or restricted passes you can give to team members with specific spending limits or model access. The system saves all your settings and data securely, so nothing is lost when you restart your server. A startup founder building an AI-powered app could use this to avoid locking into a single AI provider, easily switching between models to find the best balance of cost and quality. A product manager might use the virtual key feature to hand out access to external developers, tracking exactly how much money each developer spends. It is also a great fit for a hobbyist running local models on their own computer, allowing them to interact with those local models using the same standard tools they would use for paid cloud services. Notably, the project is designed to be simple and secure by default, requiring no manual configuration to get started. It relies on Docker, a tool that packages software into isolated units, making the gateway easy to deploy on any machine. For those exposing their gateway to the public internet, the documentation strongly recommends adding an encrypted HTTPS layer for safety.