codepilot.el

AI-powered code completion for emacs

Codepilot.el provides AI-based code completion for emacs in a similar way to copilot.el and GitHub Copilot but with open-source language models.

Getting started

Example configuration

straight.el

(use-package codepilot
  :straight (:host github :repo "ludvb/codepilot.el" :files ("*.el") )
  :hook ((prog-mode . codepilot-mode)
         (text-mode . codepilot-mode))

  :bind (:map evil-insert-state-map
              ("M-j" . 'codepilot-accept-line)
              ("M-l" . 'codepilot-accept-char)
              ("M-w" . 'codepilot-accept-word)
              ("M-<return>" . 'codepilot-accept-all)
              ("M-J" . 'codepilot-complete-next)
              ("M-K" . 'codepilot-complete-previous))

  :config
  ;; For llama.cpp backend:
  (require 'codepilot-llamacpp)
  (setq codepilot-backend 'llamacpp)
  (setq codepilot-llamacpp-address "http://localhost:8080")

  ;; For Ollama backend:
  (require 'codepilot-ollama)
  (setq codepilot-backend 'ollama)
  (setq codepilot-ollama-address "http://localhost:11434")")
  ;; Optionally, start Ollama server from emacs:
  (require 'codepilot-ollama-server)
  (codepilot-ollama-start))

Doom Emacs

;; packages.el
(package! codepilot
  :recipe (:host github :repo "ludvb/codepilot.el" :files ("*.el")))

;; config.el
(use-package! codepilot
  :hook ((prog-mode . codepilot-mode)
         (text-mode . codepilot-mode))

  :bind (:map evil-insert-state-map
              ("M-j" . 'codepilot-accept-line)
              ("M-l" . 'codepilot-accept-char)
              ("M-w" . 'codepilot-accept-word)
              ("M-<return>" . 'codepilot-accept-all)
              ("M-J" . 'codepilot-complete-next)
              ("M-K" . 'codepilot-complete-previous))

  :config
  ;; For llama.cpp backend:
  (require 'codepilot-llamacpp)
  (setq codepilot-backend 'llamacpp)
  (setq codepilot-llamacpp-address "http://localhost:8080")

  ;; For Ollama backend:
  (require 'codepilot-ollama)
  (setq codepilot-backend 'ollama)
  (setq codepilot-ollama-address "http://localhost:11434")")
  ;; Optionally, start Ollama server from emacs:
  (require 'codepilot-ollama-server)
  (codepilot-ollama-start))

Backends

Codepilot.el requires one of the following backends, which needs to be installed separately:

llama.cpp

Requirements:

• llama.cpp or a wrapper, such as llama-cpp-python
• cURL

The default prompt is designed for the DeepSeek Coder base models, which can be found on Hugging Face. If you are using another model, you may need to customize codepilot-prompt-fun and codepilot-postprocess-fun.

Ollama

Requirements:

• Ollama
• cURL

By default, the Ollama backend uses the deepseek-coder:6.7b-base model, which needs to be pulled before use by running

ollama pull deepseek-coder:6.7b-base

To use another model, customize codepilot-ollama-model. You may also need to customize codepilot-prompt-fun and codepilot-postprocess-fun as needed.

Contributing

Contributions are always welcome! Feel free to open an issue if you have found a bug or have a feature request, or submit a PR that implements a proposed change.