diff --git a/modules/age-secrets.nix b/modules/age-secrets.nix
index 2effde8..63c612d 100644
--- a/modules/age-secrets.nix
+++ b/modules/age-secrets.nix
@@ -159,5 +159,13 @@
       owner = "gitea-runner";
       group = "gitea-runner";
     };
+
+    # llama-cpp API key for bearer token auth
+    llama-cpp-api-key = {
+      file = ../secrets/llama-cpp-api-key.age;
+      mode = "0400";
+      owner = "root";
+      group = "root";
+    };
   };
 }
diff --git a/secrets/llama-cpp-api-key.age b/secrets/llama-cpp-api-key.age
new file mode 100644
index 0000000..354f211
Binary files /dev/null and b/secrets/llama-cpp-api-key.age differ
diff --git a/services/llama-cpp.nix b/services/llama-cpp.nix
index 22a766e..e1e6337 100644
--- a/services/llama-cpp.nix
+++ b/services/llama-cpp.nix
@@ -29,14 +29,18 @@
       "turbo4"
       "-fa"
       "on"
+      "--api-key-file"
+      config.age.secrets.llama-cpp-api-key.path
     ];
   };
 
   # have to do this in order to get vulkan to work
   systemd.services.llama-cpp.serviceConfig.DynamicUser = lib.mkForce false;
 
+  # Auth handled by llama-cpp --api-key-file (Bearer token).
+  # No caddy_auth — the API key is the auth layer, and caddy_auth's basic
+  # auth would block Bearer-only clients like oh-my-pi.
   services.caddy.virtualHosts."llm.${service_configs.https.domain}".extraConfig = ''
-    import ${config.age.secrets.caddy_auth.path}
     reverse_proxy :${toString config.services.llama-cpp.port}
   '';
 }