feat(config): add llama.cpp as gptel backend

configs/init-ai.el

@@ -13,8 +13,12 @@
 (use-package gptel
   :defer nil
   :config
-  (setq gptel-model 'gemma4:e2b
-        gptel-backend (gptel-make-ollama "Ollama"
+  (when (getenv "GEMINI_KEY")
+    (setq gptel-backend (gptel-make-gemini "Gemini"
+                          :key (getenv "GEMINI_KEY")
+                          :stream t)))
+
+  (setq gptel-backend (gptel-make-ollama "Ollama"
                         :host "localhost:11434"
                         :stream t
                         :request-params '(:think :json-false)
@@ -37,11 +41,16 @@
                                   "qwen3.5:2b"
                                   "qwen3.5:4b"
                                   "su_robin/gemma-4-E4B-it-Q4_K_M:latest")))
-  (when (getenv "GEMINI_KEY")
-    (setq gptel-model 'gemini-2.5-flash
-          gptel-backend (gptel-make-gemini "Gemini"
-                          :key (getenv "GEMINI_KEY")
-                          :stream t)))
+
+  (setq gptel-model 'gemma-4-e2b
+        gptel-backend
+        (gptel-make-openai "llama-cpp"
+          :stream t
+          :protocol "http"
+          :host "127.0.0.1:1945"
+          :models '("gemma-4-e4b"
+                    "gemma-4-e2b")))
+
   (setq gptel-default-mode 'org-mode
         gptel-prompt-prefix-alist
         '((markdown-mode . "# User\n\n")
@@ -96,6 +105,7 @@
   :hook (magit-mode . gptel-magit-install))
 
 (use-package gptel-autocomplete
+  :defer nil
   :ensure (:host github :repo "JDNdeveloper/gptel-autocomplete")
   :bind (("M-<return>" . gptel-complete)
          :map gptel-autocomplete-completion-map