vLLM

vLLM belongs in the database because it has become one of the most important open-source inference engines for teams serving large language models in production. The official GitHub project describes it as a fast and easy-to-use library for LLM inference and serving, and the feature list is exactly what infrastructure buyers care about: high serving throughput, PagedAttention for memory efficiency, continuous batching, OpenAI-compatible APIs, quantization support, prefix caching, and broad hardware coverage. That makes vLLM a core builder tool rather than a wrapper library.

It also deserves inclusion because it fills a real stack-layer gap. A lot of AI-builder tooling focuses on orchestration, prompting, or evaluation, but production teams still need the model-serving layer underneath. vLLM is one of the clearest answers to that problem for open models. Its pricing is also easy to frame honestly: the upstream engine is open source and free, while the real cost comes from the compute environment and any commercial hosting or support chosen around it.