Open-source LLMs for local and self-hosted use.

Three layers.

Local LLM stacks usually have a runtime, a chat or Gradio UI, and optionally a gateway. Mixing all three into one install is how machines thrash.

Ollama or llama.cpp for the runtime. Open WebUI or text-generation-webui for the shell. LiteLLM when you need one OpenAI-shaped client across several backends.

When to leave the laptop path.

vLLM is for GPU serving and throughput. It is the wrong first install on an 8 GB laptop. text-generation-webui is for loader and LoRA experiments, not a quiet daily chat box.