nixos/hosts/architect/llm.nix

{ config, pkgs, ... }:

let
  domain = "pino.giugl.io";
  backendPort = 8080;
  frontendPort = 3030;
  # llama-cpp = pkgs.unstablePkgs.llama-cpp.override { cudaSupport = true; };
  # ollama = pkgs.unstablePkgs.ollama.override { inherit llama-cpp; };
in
{
  # environment.systemPackages = [ ollama ];
  architect.vhost.${domain} = {
    dnsInterfaces = [ "tailscale" ];

    locations."/" = {
      host = "172.17.0.1";
      port = frontendPort;
      allowLan = true;
      allowWAN = true;
      # allow = [ config.architect.networks."tailscale".net ];
      extraConfig = ''
        proxy_read_timeout 600s;
      '';
    };
  };

  architect.vhost."ollama.giugl.io" = {
    dnsInterfaces = [ "tailscale" ];

    locations."/" = {
      host = "172.17.0.1";
      port = 11434;
      allowLan = true;
      allowWAN = true;
      # allow = [ config.architect.networks."tailscale".net ];
      extraConfig = ''
        proxy_read_timeout 600s;
      '';
    };
  };

  virtualisation.oci-containers = {
    containers = {
      ollama-webui = {
        image = "ghcr.io/open-webui/open-webui:main";
        autoStart = true;

        ports = [
          "172.17.0.1:${toString frontendPort}:${toString backendPort}"
        ];

        environment = {
          PORT = "${toString backendPort}";
          OLLAMA_API_BASE_URL = "http://172.17.0.1:11434/api";
        };

        extraOptions = [
          "--pull=always"
        ];
        volumes = [
          "/var/lib/ollama-webui:/app/backend/data"
        ];
      };

      ollama = {
        image = "ollama/ollama:latest";
        autoStart = true;
        extraOptions = [
          "--pull=always"
          "--gpus=all"
        ];
        environment = {
          OLLAMA_ORIGINS = "*";
        };
        volumes = [
          "/ollama:/root/.ollama"
        ];
        ports = [
          "127.0.0.1:11434:11434"
          "172.17.0.1:11434:11434"
        ];
      };
    };
  };
}