Local model runtimes

What it is

A local runtime runs compatible model files on your computer or a server you control. Ollama supplies a model API; LM Studio adds a desktop interface and document chat. Neither a downloaded model nor a localhost address alone proves that the complete workflow stays on the device.

How it works

Choose a model and record its exact identifier, quantization and runtime version. Check that the hardware can load it and that its usable context fits the task. Prepare model files before disconnecting. In LM Studio, attach a supported document; for a custom Ollama application, supply your own document preparation and retrieval layer. Test the path with non-sensitive sample data first.

Benefits

Local execution can make a small knowledge experiment possible without sending prompts to a hosted model. It also lets a team keep a fixed model artifact for comparisons. Start with known-answer questions and save the configuration beside the results so another person can repeat the same test.

Limits and risks

Cloud models, remote MCP servers, plugins and synchronized output folders can change where data goes. Ollama documents a switch to disable its cloud features. Review network connections and endpoint access separately. Local models can still invent answers, miss relevant text or expose data to another user of the machine. Model licences and hardware costs also remain relevant.