Use Ollama, LM Studio or other local servers
isoboard can send captures to any server that speaks the OpenAI chat API, including Ollama and LM Studio. Enter the server’s endpoint URL on the Providers tab, then type the ID of a vision-capable model in a Profile. A server on your own Mac keeps images on your device.
How do I connect an OpenAI-compatible server?
Section titled “How do I connect an OpenAI-compatible server?”- Start your server and load a model that accepts images.
- In isoboard, open Settings and select the Providers tab.
- Expand OpenAI-compatible.
- Enter the server’s address in Endpoint, ending in
/v1. Ollama typically listens athttp://localhost:11434/v1and LM Studio athttp://localhost:1234/v1. - If your server requires a key, enter it in API key (optional). Leave the field empty for servers that need none, such as a default Ollama or LM Studio install.
- Open the Profiles tab, set Provider to OpenAI-compatible, and type the model ID exactly as the server lists it.
isoboard has no bundled model list for this provider, so you always enter the model ID yourself.
What endpoint rules apply?
Section titled “What endpoint rules apply?”http://works only forlocalhost,127.0.0.1and[::1]. Any other host must usehttps://.
A server on another computer on your network needs HTTPS to work.
Which models work?
Section titled “Which models work?”The model must accept images, because isoboard sends the capture as an image. A text-only model fails at capture time. isoboard does not block it when you save. A text-only model can still serve the separate Additional Rules step if you pick it for that step in a Profile.
Is it private?
Section titled “Is it private?”It depends on where the server runs. A server on your own Mac keeps the image on your device, similar to a local model. A remote server receives the image, and its operator applies its own policies. See what leaves your Mac.
Where is my API key stored?
Section titled “Where is my API key stored?”In the macOS Keychain, never in the configuration file. See where isoboard stores your data.
Troubleshooting
Section titled “Troubleshooting”- Nothing happens or the request fails: check the server is running and the endpoint ends in
/v1. - The server rejects the image: the model probably has no vision support.
- Long waits: a large model may take longer than the Profile’s timeout, which defaults to 120 seconds. Raise it in the Profile’s advanced options.
See also provider errors.