Show HN: Small hardware box that runs local LLMs and exposes an OpenAI API

I’ve been building a small hardware box that runs local LLMs like Mistral, Qwen and Llama, and exposes an OpenAI compatible API on your local network. There’s no cloud, no login system and no telemetry. I built it because a lot of small firms want ChatGPTbstyle tools but can’t use cloud AI for privacy or compliance reasons, and most don’t want to deal with GPU servers, drivers, Docker or model configs. The aim is to make local AI feel as simple as plugging in a router.

Right now the box boots into a very simple web UI where you choose a model and start using it. The API follows the OpenAI format for chat completions and embeddings. It can run different models depending on the hardware you pick, either a Jetson Orin Nano or an x86 mini-PC with a GPU. It stores data locally, supports basic RAG indexing and only exposes itself on the LAN by default.

A few things still aren’t working. There’s no multi-user rate limiting yet. The RAG quality is basic and I’m still improving chunking and reranking. The Orin runs hot under heavy load, so thermal performance needs work. It’s also still a prototype rather than a finished consumer product.

On the technical side, it runs containerized model servers using Ollama and some custom runners. Models load through GGUF or TensorRT-LLM depending on the hardware. The API layer follows the OpenAI spec. The RAG pipeline uses local embeddings and a vector database. The software stack is a mix of TypeScript and Python.

I’m looking for feedback from anyone who has built or deployed local inference before. I’m trying to understand what thermal and power issues you’ve run into, whether a drop-in OpenAI compatible box is actually useful to small teams, what hardware setups I should consider, and any honest critiques of the idea.