vLLM
High-throughput LLM inference and serving engine with an OpenAI-compatible server for NVIDIA, AMD, Intel, TPU and CPU.
- License
- Apache-2.0
- Platforms
- Linux, Docker, CLI
- Own API keys
- Yes
- Checked
- 2026-09-24
vLLM works with OpenAI-compatible APIs. Grab a free key and a model from the catalog: for coding · for agents · providers and base URLs
vLLM alternatives
Popular runtime to download and run open LLMs locally via CLI, desktop app and a REST/OpenAI-compatible API.
Desktop app to discover, download and run local LLMs (llama.cpp, MLX) with a chat UI and an OpenAI-compatible local server.
Offline ChatGPT alternative: runs local LLMs, connects to cloud APIs, supports MCP and serves an OpenAI-compatible API on localhost.
Reference C/C++ engine for GGUF models; llama-server provides an OpenAI-compatible API and web UI on CPU, CUDA, Metal, Vulkan and ROCm.
Self-hosted drop-in OpenAI/Anthropic API replacement for LLMs, images, audio, video and embeddings on CPU or GPU.
Single-file llama.cpp-based runner with the KoboldAI Lite UI; OpenAI/Ollama-compatible APIs plus image generation, Whisper and TTS.
FAQ
Is vLLM free?
Completely free and open source, but it needs your own hardware, usually a server GPU.
Does vLLM work with free models?
Yes. It supports your own keys and OpenAI-compatible APIs, so you can plug in free models from OpenRouter, Groq, NVIDIA Build, or a local Ollama.
What license does vLLM use?
Apache-2.0. The source code is open.