Show HN: GPUWorks – A 40 MB GPU cluster manager for vLLM/SGLang, no K8s needed

Category: infrastructure

Tags: gpu-cluster, model-serving, vllm, sglang, heterogeneous-compute, go, maas, openai-compatible

Score: 7.5/10 (Innovation: 7, Technical: 8, Documentation: 8, Utility: 7)

GPUWorks is a lightweight Go-based control plane for managing heterogeneous GPU clusters and serving models via vLLM/SGLang with an OpenAI-compatible gateway, eliminating Kubernetes for many MaaS deployments. It is notable for its 40 MB single-binary footprint, multi-vendor accelerator support (NVIDIA, AMD, Ascend, PPU, Hygon, Apple Silicon), and verified model×accelerator launch specs.

Target audience: DevOps, ML platform engineers, infrastructure engineers, and backend developers deploying self-hosted LLM inference

Repository: https://github.com/flitai-io/gpuworks · Go · Apache-2.0 · 60 stars

View on Hacker News