LLM X

Privacy statement:

LLM X does not make any external api calls. (go ahead, check your network tab and see the Fetch section). Your chats and image generations are 100% private. This site / app works completely offline.

Issues:

LLM X (web app) will not connect to a server that is not secure. This means that you can use LLM X on localhost (considered a secure context) but if you're trying to use llm-x over a network the server needs to be from https or else it will not work.

Recent additions:

Multi-model support! Chats now support multiple models, from any connected server, and can respond with each them at the same time
Quick Delete! using the quick search/ kbar panel, it is super easy to delete all app related data
Chrome extension support! All the features of the web app, built into the browser
Gemini Nano support!
IndexedDB support! All text is now saved in IndexedDB instead of local storage
Auto-connect on load. If there is a server thats visible and ready to connect, we connect to it for you

How To Use:

Prerequisites for application

Ollama: Download and install Ollama
- Pull down a model (or a few) from the library Ex: ollama pull llava (or use the app)
LM Studio: Download and install LM Studio
AUTOMATIC1111: Git clone AUTOMATIC1111 (for image generation)
Gemini Nano: Download and install Chrome Canary
- Enable On Device Model by selecting BypassPerfRequirement
- Enable Api Gemini for nano
- Relaunch Chrome (may need to wait for it to download)

How to use web client (no install):

Prerequisites for web client

Ollama Options:
- Use Ollama's FAQ to set OLLAMA_ORIGINS = https://mrdjohnson.github.io
- Run this in your terminal OLLAMA_ORIGINS=https://mrdjohnson.github.io ollama serve
  - (Powershell users: $env:OLLAMA_ORIGINS="https://mrdjohnson.github.io"; ollama serve)
LM Studio:
- Run this in your terminal: lms server start --cors=true
A1111:
- Run this in the a1111 project folder: ./webui.sh --api --listen --cors-allow-origins "*"
Gemini Nano: works automatically

Use your browser to go to LLM-X
Go offline! (optional)
Start chatting!

Prerequisites for chrome extension

Download and install Chrome Extension
Ideally this works out of the box, no special anything needed to get it connecting! If not continue with the steps below:
Ollama Options:
- Use Ollama's FAQ to set OLLAMA_ORIGINS = chrome-extension://iodcdhcpahifeligoegcmcdibdkffclk
- Run this in your terminal OLLAMA_ORIGINS=chrome-extension://iodcdhcpahifeligoegcmcdibdkffclk ollama serve
  - (Powershell users: $env:OLLAMA_ORIGINS="chrome-extension://iodcdhcpahifeligoegcmcdibdkffclk"; ollama serve)
LM Studio:
- Run this in your terminal: lms server start --cors=true
A1111:
- Run this in the a1111 project folder: ./webui.sh --api --listen --cors-allow-origins "*"

How to use offline:

Follow instructions for "How to use web client"
In your browser search bar, there should be a download/install button, press that.
Go offline! (optional)
Start chatting!

How to use from project source:

Prerequisites for project source

Ollama: Run this in your terminal ollama serve
LM Studio: Run this in your terminal: lms server start
A1111: Run this in the a1111 project folder: ./webui.sh --api --listen

Vite preview mode

Pull down this project; yarn install, yarn preview
Go offline! (optional)
Start chatting!

Docker

Run this in your terminal: docker compose up -d
Open http://localhost:3030
Go offline! (optional)
Start chatting!

Chrome Extension

Pull down this project; yarn chrome:build
Navigate to chrome://extensions/
Load unpacked (developer mode option) from path: llm-x/extensions/chrome/dist

Goals / Features

Screenshots:

Showing Chrome extension mode with Google's on-device Gemini Nano

Showing Chrome extension mode with Ollama's llama3.2-vision

Showing ability to run ollama and LM Studio at the same time

Conversation about logo

Image generation example!

Showing off omnibar and code

Showing off code and light theme

Responding about a cat

LaTex support!
$Latex screenshot$

Another logo response

What is this? ChatGPT style UI for the niche group of folks who run Ollama (think of this like an offline chat gpt server) locally. Supports sending and receiving images and text! WORKS OFFLINE through PWA (Progressive Web App) standards (its not dead!)

Why do this? I have been interested in LLM UI for a while now and this seemed like a good intro application. I've been introduced to a lot of modern technologies thanks to this project as well, its been fun!

Why so many buzz words? I couldn't help but bee cool 😎

Tech Stack (thank you's):

Logic helpers:

React
Typescript
Lodash
Zod
Mobx
~~Mobx State Tree~~ (great for rapid starting a project, terrible for long term maintenance)
Langchain.js

UI Helpers:

Project setup helpers:

Inspiration: ollama-ui's project. Which allows users to connect to ollama via a web app

Perplexity.ai Perplexity has some amazing UI advancements in the LLM UI space and I have been very interested in getting to that point. Hopefully this starter project lets me get closer to doing something similar!

Getting started with local development

(please note the minimum engine requirements in the package json)

Clone the project, and run yarn in the root directory

yarn dev starts a local instance and opens up a browser tab under https:// (for PWA reasons)

MISC

LangChain.js was attempted while spiking on this app but unfortunately it was not set up correctly for stopping incoming streams, I hope this gets fixed later in the future OR if possible a custom LLM Agent can be utilized in order to use LangChain
- edit: Langchain is working and added to the app now!
Originally I used create-react-app 👴 while making this project without knowing it is no longer maintained, I am now using Vite. 🤞 This already allows me to use libs like ollama-js that I could not use before. Will be testing more with langchain very soon
This readme was written with https://stackedit.io/app
Changes to the main branch trigger an immediate deploy to https://mrdjohnson.github.io/llm-x/

mrdjohnson/llm-x