Most local apps can run a small web server on your own computer that accepts the same requests as OpenAI’s API. Anything built to talk to that API (editors, note apps, scripts, agents) can then use your local model by changing two settings: the base URL and the model name.
| App | Default address | How to start it |
|---|---|---|
| Ollama | http://localhost:11434/v1 | Runs in the background once installed. |
| LM Studio | http://localhost:1234/v1 | Start the local server from the Developer tab. |
| llama.cpp | http://localhost:8080/v1 | llama-server -m model.gguf |
Pointing a tool at it
- Load or pull the model in the app and start its server.
- In the other tool, choose an OpenAI-compatible or “custom” provider.
- Set the base URL to the address above. Use the model name exactly as the app lists it.
- For the API key, enter any placeholder text if the tool insists on one. Local servers normally ignore it.
Things worth knowing
- Context. The app sets how much context the model gets, not the tool. If answers cut off or forget, raise the context length in the app (memory cost).
- Tool calling and images. Whether they work depends on the model and the app version, not just the API.
- Keep it on your machine. By default these servers listen only on your own computer. Do not expose one to a network without a password or proxy in front of it. See privacy.
- One model at a time. Switching models reloads weights, which takes a few seconds, and keeps both in memory if it fits.
Pick an app on the apps page or read which app to use.