Your finish line
A local model answers a harmless prompt on your own computer, with a clear idea of what stays local and what could leave it.
- Check your hardware. Start with a smaller quantized model and a short context rather than filling every last gigabyte.
- Choose LM Studio for a desktop interface, or Ollama for a command-line/API foundation. You do not need both.
- Send a test question and inspect memory use. If loading fails, use the memory checklist.
- With Ollama working, try the local chat-interface project. If documents are your goal, use the document assistant project instead.
Know what local means
A local interface can still call a cloud model, remote embeddings or web-search tool. Check each provider and connector. Downloads and updates normally need a connection, even when inference can later run offline.
Read for perspective
Open versus closed models: choosing for a workflow is archived April 2026 analysis. Use it for the tradeoffs, not as a current price or model list.