How to access remote Ollama server domain

[clickbrain]

I have ollama hosted on a Mac Studio and have a domain mapped to the Ollama URL on that server: https://ollama.mydomain.com. Unfortunately, it looks like Goose adds :11434/v1/chat/completions to the end of whatever host a user puts into the Ollama setup field. Is there anyway to make sure it doesn't do that and just maps to the endpoint of the URL I provided?

I am assuming that is why I am getting the "Ran into this error: Execution error: error sending request for url" message, because I have Qwen 2.5 in Ollama on the remote server.

[wendytang]

Would it be possible to set up a reverse proxy? Like

server {
    listen 443 ssl;
    server_name ollama.mydomain.com;

    location / {
        proxy_pass http://127.0.0.1:11434;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    }
}

[cbruyndoncx]

maybe adding the :80 port is enough for goose to not do that ?

[Richie765]

I'm running into the same problem.

I have Ollama running on https://ollama.mydomain.com.

When I entered this URL in Goose (Desktop), it would try to access it on port 11434 instead of 443. After some effort, I added the url with the :443 port and now I get this error:

Ran into this error: Execution error: error sending request for url (https://ollama.mydomain.com/v1/chat/completions).

When I click on this link, it goes to the browser and this URL does not exist.

I looked into the Ollama API documentation at https://github.com/ollama/ollama/blob/main/docs/api.md and there all URL's begin with /api/ and there is nothing that ends with /chat/completions or /completions

Also I noticed that if I change the URL within Goose, I always have to Quit and Restart Goose before it takes effect.

[NerdyShawn]

Just to add to this defining the port ollama host did override the 11434 default ok for me, thanks! I'm using the open-webui chart which deploys an ollama sub chart as well accessible in k8s cluster via ingress.

🟢

$ goose configure

Welcome to goose! Let's get you set up with a provider.
  you can rerun this command later to update your configuration

┌   goose-configure 
│
◇  Which model provider should we use?
│  Ollama 
│
◇  Provider Ollama requires OLLAMA_HOST, please enter a value
│  https://ollama.somecooldomain.lan:443
│
◇  Model fetch complete
│
◇  Enter a model from that provider:
│  devstral:latest
│
◓  Checking your configuration...                                                                                                                                                       
└  Configuration saved successfully

  Tip: Run 'goose configure' again to adjust your config or add extensions

[DOsinga]

So we should default to 443 if we have a fully qualified url?

[Richie765]

I'd say, default port for http is 80, for https it is 443. This is just the standard and everybody knows it. If you need another port you specify it.

[cbruyndoncx]

if custom providers are fully functional, this is a good test case ...

[DOsinga]

we have custom providers now! give it a whirl!