{
  pkgs,
  service_configs,
  config,
  inputs,
  lib,
  ...
}:
{
  services.llama-cpp = {
    enable = true;
    model = builtins.toString (
      pkgs.fetchurl {
        url = "https://huggingface.co/ggml-org/gpt-oss-20b-GGUF/resolve/main/gpt-oss-20b-mxfp4.gguf";
        sha256 = "52f57ab7d3df3ba9173827c1c6832e73375553a846f3e32b49f1ae2daad688d4";
      }
    );
    port = service_configs.ports.llama_cpp;
    host = "0.0.0.0";
    # vulkan broken: https://github.com/ggml-org/llama.cpp/issues/13801
    package = (lib.optimizePackage inputs.llamacpp.packages.${pkgs.system}.default);
    extraFlags = [
      # "-ngl"
      # "9999"
    ];
  };

  # have to do this in order to get vulkan to work
  systemd.services.llama-cpp.serviceConfig.DynamicUser = lib.mkForce false;

  services.caddy.virtualHosts."llm.${service_configs.https.domain}".extraConfig = ''
    ${builtins.readFile ../secrets/caddy_auth}
    reverse_proxy :${builtins.toString config.services.llama-cpp.port}
  '';
}