vLLM
Open-source engine for serving LLMs at high throughput on your own GPUs, with an OpenAI-compatible API server; Apache-2.0.
vLLM (v0.32.0, Oct 2026, ~95.6k stars) is an inference and serving engine for large language models. It uses PagedAttention and continuous batching to serve many requests on your own GPUs and exposes an OpenAI-compatible API server. Install with uv pip install vllm, then vllm serve a model. Apache-2.0, no paid services; each model keeps its own licence. Needs Linux and an NVIDIA GPU with compute capability 7.5+ (or a CPU or other accelerator backend); macOS on Apple silicon is experimental and there is no native Windows (use WSL). Memory needs depend on the model.
The facts
| Question | Answer | Source | Checked |
|---|---|---|---|
| Runs on my machine? | Yes. Yes, you install it with pip/uv and serve models on your own machine or server | github.com | Checked 12 days ago |
| Code licence | Open source. Apache-2.0 | github.com | Checked 13 days ago |
| Commercial use | Yes. Engine is Apache-2.0; commercial use of a model depends on that model's licence | github.com | Checked 12 days ago |
| Paid services needed? | None required; runs open models on your own hardware | github.com | Checked 12 days ago |
| Hardware | NVIDIA GPUs need compute capability 7.5+ (T4, RTX 20xx, A100 and newer); CPU backends exist for x86, ARM, Apple silicon and IBM Z; no minimum RAM or VRAM figures, as memory depends on the model | docs.vllm.ai | Checked 12 days ago |
| Operating systems | Linux is the supported OS; macOS on Apple silicon is experimental and build-from-source only; no native Windows (use WSL) | docs.vllm.ai | Checked 12 days ago |
| Setup effort | Medium; one pip/uv install, but needs Linux, a supported GPU or CPU build, and model downloads | github.com | Checked 12 days ago |
Each fact has its own date. Unknown means nobody has published it. We read their documentation; we didn’t install it. Running it yourself costs your hardware, electricity and setup time. Report a correction.
No licence changes since we started watching. Licence changes: follow vLLM and we email you if its terms change.
Stars over time
Building history +537 this week; not enough baseline history. Stars show attention, not quality. Where these numbers come from.