Add llama-cpp and gptel

3 files changed, 57 insertions, 3 deletions

diff --git a/.config/emacs/init.el b/.config/emacs/init.el
index b8d911b..c46441a 100644
--- a/.config/emacs/init.el
+++ b/.config/emacs/init.el
@@ -103,7 +103,7 @@
 (require 'bd--project)
 (require 'bd--files)
 (require 'bd--dictionary)
-(require 'bd--irc)
+(require 'bd--chat)
 (require 'bd--shells)
 (require 'bd--minibuffer)
 (require 'bd--buffer)
diff --git a/.config/emacs/modules/bd--irc.el b/.config/emacs/modules/bd--chat.el
index fe2d58f..a55f10a 100644
--- a/.config/emacs/modules/bd--irc.el
+++ b/.config/emacs/modules/bd--chat.el
@@ -58,6 +58,56 @@ channel, which is issuing the PART command."
              :port 6697
              :encryption tls))))
 
+(use-package gptel
+  :bind (("C-c g" . gptel-menu))
+  :config
+  (defvar bd/llama-cpp-buffer-name "*llama-cpp-proc*")
+  (defvar bd/llama-cpp-reasoning-buffer-name "*llama-cpp-reasoning*")
+  (defvar bd/llama-cpp-port "4568")
+  (defvar bd/llama-cpp-threads "8")
+  (defvar bd/llama-cpp-model-file "~/.config/guix/assets/Qwen3-4B.Q3_K_M.gguf")
+  (defvar bd/llama-cpp-reasoning-budget nil)
+  (defun bd/gptel-start-backend ()
+    (interactive)
+    (let ((process (get-buffer-process bd/llama-cpp-buffer-name)))
+      (if process
+	  (message "llama-cpp process is already running!")
+	(progn
+	  (start-process-shell-command
+	   "llama-cpp" bd/llama-cpp-buffer-name
+	   (concat "llama-server --reasoning-budget "
+		   (if bd/llama-cpp-reasoning-budget "-1" "0")
+		   " --port " bd/llama-cpp-port
+		   " -t " bd/llama-cpp-threads
+		   " -m " bd/llama-cpp-model-file)))
+	(unless (get-buffer bd/llama-cpp-reasoning-buffer-name)
+	  (generate-new-buffer bd/llama-cpp-reasoning-buffer-name)))))
+  (defun bd/gptel-stop-backend ()
+    (interactive)
+    (let ((process (get-buffer-process bd/llama-cpp-buffer-name)))
+      (if process
+	  (progn
+	    (delete-process process)
+	    (kill-buffer bd/llama-cpp-buffer-name)
+	    (message "Killed %s." process))
+	(message "No llama-cpp process is running."))))
+  (defun bd/gptel-restart-backend ()
+    (interactive)
+    (bd/gptel-stop-backend)
+    (bd/gptel-start-backend))
+
+  (setopt gptel-model 'qwen-4b
+	  gptel-backend (gptel-make-openai "llama-cpp"
+			  :stream t
+			  :protocol "http"
+			  :host (concat "localhost:" bd/llama-cpp-port)
+			  :models '(qwen-4b))
+	  gptel-max-tokens 500
+	  gptel-include-reasoning bd/llama-cpp-reasoning-buffer-name)
+
+  (gptel-make-preset 'default
+    :system "You are a wolf (furry) named Evka hired as a secretary to complete language-based tasks. First describe an action your character does, e.x.: *I tap my claws on the desk*. Finish by responding to the task as tersely as possible, in character."))
+
 
-(provide 'bd--irc)
-;;; bd--irc.el ends here
+(provide 'bd--chat)
+;;; bd--chat.el ends here
diff --git a/.config/guix/modules/services/emacs.scm b/.config/guix/modules/services/emacs.scm
index 9c42906..fa70880 100644
--- a/.config/guix/modules/services/emacs.scm
+++ b/.config/guix/modules/services/emacs.scm
@@ -21,6 +21,7 @@
  cpp
  gnupg
  python-xyz
+ machine-learning
  mail
  lisp)
 
@@ -39,6 +40,7 @@
    emacs-emms
    emacs-exwm
    emacs-f
+   emacs-gptel
    emacs-guix
    emacs-hydra
    emacs-jeison
@@ -57,6 +59,8 @@
 
    ccls                ;; for eglot
    python-lsp-server   ;;
+
+   llama-cpp           ;; for emacs-gptel
    ))