Skip to content

Latest commit

 

History

History
168 lines (121 loc) · 5.06 KB

README.md

File metadata and controls

168 lines (121 loc) · 5.06 KB
Raw

A snappy, keyboard-centric terminal user interface for interacting with large language models.
Chat with Claude 3, ChatGPT, and local models like Llama 3, Phi 3, Mistral and Gemma.

elia-screenshot-collage

Introduction

elia is an application for interacting with LLMs which runs entirely in your terminal, and is designed to be keyboard-focused, efficient, and fun to use! It stores your conversations in a local SQLite database, and allows you to interact with a variety of models. Speak with proprietary models such as ChatGPT and Claude, or with local models running through ollama or LocalAI.

Installation

Install Elia with pipx:

pipx install --python 3.11 elia-chat

Depending on the model you wish to use, you may need to set one or more environment variables (e.g. OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY etc).

Quickstart

Launch Elia from the command line:

elia

Launch a new chat inline (under your prompt) with -i/--inline:

elia -i "What is the Zen of Python?"

Launch a new chat in full-screen mode:

elia "Tell me a cool fact about lizards!"

Specify a model via the command line using -m/--model:

elia -m gpt-4o

Options can be combined - here's how you launch a chat with Gemini 1.5 Flash in inline mode (requires GEMINI_API_KEY environment variable).

elia -i -m gemini/gemini-1.5-flash-latest "How do I call Rust code from Python?"

Running local models

  1. Install ollama.
  2. Pull the model you require, e.g. ollama pull llama3.
  3. Run the local ollama server: ollama serve.
  4. Add the model to the config file (see below).

Configuration

The location of the configuration file is noted at the bottom of the options window (ctrl+o).

The example file below shows the available options, as well as examples of how to add new models.

# the ID or name of the model that is selected by default on launch
default_model = "gpt-4o"
# the system prompt on launch
system_prompt = "You are a helpful assistant who talks like a pirate."

# choose from "nebula", "cobalt", "twilight", "hacker", "alpine", "galaxy", "nautilus", "monokai", "textual"
theme = "galaxy"

# change the syntax highlighting theme of code in messages
# choose from https://pygments.org/styles/
# defaults to "monokai"
message_code_theme = "dracula"

# example of adding local llama3 support
# only the `name` field is required here.
[[models]]
name = "ollama/llama3"

# example of a model running on a local server, e.g. LocalAI
[[models]]
name = "openai/some-model"
api_base = "http://localhost:8080/v1"
api_key = "api-key-if-required"

# example of add a groq model, showing some other fields
[[models]]
name = "groq/llama2-70b-4096"
display_name = "Llama 2 70B"  # appears in UI
provider = "Groq"  # appears in UI
temperature = 1.0  # high temp = high variation in output
max_retries = 0  # number of retries on failed request

# example of multiple instances of one model, e.g. you might
# have a 'work' OpenAI org and a 'personal' org.
[[models]]
id = "work-gpt-3.5-turbo"
name = "gpt-3.5-turbo"
display_name = "GPT 3.5 Turbo (Work)"

[[models]]
id = "personal-gpt-3.5-turbo"
name = "gpt-3.5-turbo"
display_name = "GPT 3.5 Turbo (Personal)"

Custom themes

Add a custom theme YAML file to the themes directory. You can find the themes directory location by pressing ctrl+o on the home screen and looking for the Themes directory line.

Here's an example of a theme YAML file:

name: example  # use this name in your config file
primary: '#4e78c4'
secondary: '#f39c12'
accent: '#e74c3c'
background: '#0e1726'
surface: '#17202a'
error: '#e74c3c'  # error messages
success: '#2ecc71'  # success messages
warning: '#f1c40f'  # warning messages

Changing keybindings

Right now, keybinds cannot be changed. Terminals are also rather limited in what keybinds they support. For example, pressing Cmd+Enter to send a message is not possible (although we may support a protocol to allow this in some terminals in the future).

For now, I recommend you map whatever key combo you want at the terminal emulator level to send \n. Here's an example using iTerm:

image

With this mapping in place, pressing Cmd+Enter will send a message to the LLM, and pressing Enter alone will create a new line.

Import from ChatGPT

Export your conversations to a JSON file using the ChatGPT UI, then import them using the import command.

elia import 'path/to/conversations.json'

Wiping the database

elia reset

Uninstalling

pipx uninstall elia-chat