Self-hosted AI that fits your machine.
Updated Sep 11, 2026
Start from the machine.
Self-hosted AI fails when the catalog ranks stars ahead of RAM and GPU. A laptop with 8 GB and no discrete GPU is a different job from a home GPU box or a VPS with Docker.
Fossfits scores fit for that profile. Use Local AI for chat and runners. Use RAG and app builders for document workspaces. Use LLM infrastructure when you need a gateway or a serving layer.
A short path.
Ollama plus Open WebUI covers most “self-hosted ChatGPT” installs. Jan skips Docker when you want a desktop window. AnythingLLM is the document Q&A layer on top of a runner you already trust.
vLLM and Dify are the next step only when you have GPU headroom or a real multi-user workflow.
What this guide skips.
Hosted SaaS chat, closed weights you cannot download, and “best AI of 2026” listicles. Those pages do not tell you whether the stack boots on your box.