ELAIZA

An updated Emacs doctor.

This package provides a Large Language Model independent (LLM) backend with multiple applications. For example, to honor its ancestor, you can talk to a new Artificial Intelligence-enhanced (AI) Emacs doctor via M-x elaiza-doctor.

For general chatting, you can use M-x elaiza-chat directly from the minibuffer. By default, you can chat with a local Rocket-3B model (see getting started). You can switch backends by prefixing your commands with C-u and selecting an available LLM (see supported backends).

Similar to OpenAI’s GPTs, you can create your own assistants. For example, M-x elaiza-editor is a multilingual editor, giving you suggestions for your current buffer. With elaiza-jupyter you can get org-mode responses that are optimized for Python code, similar to Jupyter Notebooks.

Using ELAIZA as the backend, you can build all kind of applications, like elaiza-businesscard, which parses an image of a business card and returns the annotated, OCR’ed result in org-mode format.

Installation

Doom Emacs

Add the following to your packages.el

(package! elaiza :recipe (:host github :repo "SFTtech/emacs-elaiza" :branch "main"))

Usage

Getting Started

If you have no prior experience with LLMs and want to try ELAIZA, simply use M-x elaiza-chat. This will download a small model (Rocket-3B, 1.89 GB) to ~/llamafiles and start the the server automatically. Afterwards you can start chatting with the model. Note, due to its size, it will have worse capabilities compared to larger models, such as GPT-4. You can stop the server with M-x elaiza-llamafile-stop.

Key Bindings

In elaiza-mode, you can use C-c <RET> to continue the conversation (elaiza-chat-continue) and C-c C-k to interrupt the response (elaiza-chat-interrupt).

Applications

Currently, the following applications are available:

• M-x elaiza-chat for direct minibuffer prompts
• M-x elaiza-doctor for an AI-enhanced Emacs doctor experience
• M-x elaiza-editor as a multilingual editor for suggestions in your current buffer
• M-x elaiza-jupyter like elaiza-chat but with special Python instructions.
• M-x elaiza-businesscard parse an image of a businesscard using GPT-4o mini.

Specifying a default model

You can specify a default backend using customize. Alternatively prefix the elaiza commands (C-u) or call M-x elaiza-change-default-model.

GPT-4o

For example, in Doom Emacs, insert the following into your config.el to use GPT-4o as your default model.

(use-package! elaiza
  :config (setq elaiza-default-model (make-elaiza-gpt-4o))

Llamafile: LLaMA-3 Instruct 8B

If you have downloaded a Llamafile already, for example, from https://github.com/mozilla-Ocho/llamafile, you can select it as default model by specifying its name and location.

(use-package! elaiza
  :config
  (setq elaiza-default-model (make-elaiza-llamafile
                              :name "Llamafile: LLaMA-3 8B"
                              :filename "~/llamafiles/llama3.llamafile")))

API Keys

Some backends, such as ChatGPT and Claude 3, require an API key. To securely store and retrieve API keys, use auth-source, as documented in the Emacs Auth Manual. Add the following to your auth-sources file, for example, .authinfo.gpg, to store them:

Example: OpenAI

Create a key at https://platform.openai.com/api-keys. Insert into your ~/.authinfo.gpg:

machine api.openai.com port https login elaiza password <your-api-key>

Use GPT-4 Turbo as your default model by adding the following to your config.el:

(use-package! elaiza
  :config (setq elaiza-default-model (make-elaiza-gpt-4-turbo))

Example: Claude

Create a key at https://console.anthropic.com/settings/keys. Insert into your ~/.authinfo.gpg:

machine api.anthropic.com port https login elaiza password <your-api-key>

Use Claude 3 Opus as your default model by adding the following to your config.el:

(use-package! elaiza
  :config (setq elaiza-default-model (make-elaiza-claude-opus))

Supported Backends

Model	Provider	Sourcecode
GPT 4o	OpenAI	elaiza-openai.el
GPT 4o mini	OpenAI	elaiza-openai.el
GPT 4	OpenAI	elaiza-openai.el
GPT 4 Turbo	OpenAI	elaiza-openai.el
GPT 3.5 Turbo	OpenAI	elaiza-openai.el
Claude 3 Opus	Anthropic	elaiza-claude.el
Claude 3 Sonnet	Anthropic	elaiza-claude.el
Claude 3 Haiku	Anthropic	elaiza-claude.el
Available Llamafiles	Llamafile	elaiza-llamafile.el
Available Ollama Models	Ollama	elaiza-ollama.el

Alternatives

• karthink/gptel GPTel: A simple LLM client for Emacs
• ahyatt/llm: llm package for emacs
• s-kostyaev/ellama: Ellama is a tool for interacting with large language models from Emacs.