PhyseaWiki How AI actually works physea.ai →

Serving & runtimes

What are the easiest ways to run a model on my own machine?

Ollama runs models from the command line, pulling them by name like Docker images. LM Studio is a point-and-click desktop app with a built-in model browser and chat. Both give you a local OpenAI-compatible server.

Last updated 2026-07-25 · Physea Labs

If you want to get a model running today, these two tools ask the least of you.

Ollama is the command-line option. You pull a model by name and run it, much the way you would pull a Docker image. The project’s own tagline lists models you can start with, such as Qwen, Gemma, DeepSeek, and Llama.[1] It is open source under the MIT license,[1] gives you a local REST API, and offers OpenAI-compatible chat completions so other apps can talk to it.[2] Note one current limit: Ollama’s OpenAI-compatible layer covers chat completions, and an embeddings API was listed as still under consideration rather than available.[2]

What you pull is more than a raw weights file. Every entry in Ollama’s library is built from a Modelfile, a small blueprint format that bundles the base weights with a prompt template, default generation parameters, and an optional system message.[4] That is why the same base model can behave a little differently through two different tools even though the weights are byte-identical: the template and defaults baked into the Modelfile are not.

LM Studio is the point-and-click option. It is a desktop application for running models locally, with a built-in browser for finding and downloading them and a chat window for talking to them.[3] When you are ready to wire it into your own code, it can serve OpenAI-like endpoints on your machine and across your network.[3] It also offers a headless mode for servers or machines where you do not want a GUI, useful for a Mac mini or spare PC left running so your phone or another device on the network can reach it without a monitor attached.[3]

A useful thing to know: LM Studio runs llama.cpp under the hood for GGUF models, and adds Apple’s MLX engine on Apple Silicon.[3] So the friendly window you click in is sitting on top of the same engine covered on the next page.

Ollama – Command line– Pull models by name– MIT license LM Studio – Desktop window– Browse and click to download– Headless mode available
Under the interface, both end up serving the same OpenAI-compatible shape — the difference is entirely in how you get there.

Friendly local runtimes

  • Ollama ↗

    Command-line tool that pulls and runs models by name; local REST and OpenAI-compatible API.

  • LM Studio ↗

    Desktop app with a model browser, chat window, and a local OpenAI-compatible server.

References

  1. Ollama README — Ollama
  2. OpenAI compatibility — Ollama
  3. LM Studio Documentation — LM Studio
  4. Modelfile Reference — Ollama

Related