github

smol-ai / GodMode

  • понедельник, 21 августа 2023 г. в 00:00:20
https://github.com/smol-ai/GodMode

AI Chat Browser: Fast, Full webapp access to ChatGPT / Claude / Bard / Bing / Llama2! I use this 20 times a day.



🐣 GodMode - the smol AI Chat Browser

This is a dedicated chat browser that only does one thing: help you quickly access the full webapps of ChatGPT, Claude 2, Perplexity, Bing and more with a single keyboard shortcut (Cmd+Shift+G).

image

(click for video)

Whatever is typed at the bottom is entered into all web apps simultaneously, however if you wish to explore one further than the other you can do so independently since they are just webviews.

Installation

Install here! Arm64 for Apple Silicon Macs, non Arm64 for the rest. We technically do support Windows, but need Windows users to help us with the build process/instructions.

You can also build from source, see instructions below.

Mixture of Mixture of Experts

It's well discussed by now that GPT4 is a mixture of experts model, which explains its great advancement over GPT3 while not sacrificing speed. It stands to reason that if you can run one chat and get results from all the top closed/open source models, you will get that much more diversity in results for what you seek. As a side benefit, we will add opt-in data submission soon so we can crowdsource statistics on win rates, niche advantages, and show them over time.

“That's why it's always worth having a few philosophers around the place. One minute it's all is truth beauty and is beauty truth, and does a falling tree in the forest make a sound if there's no one there to hear it, and then just when you think they're going to start dribbling one of 'em says, incidentally, putting a thirty-foot parabolic reflector on a high place to shoot the rays of the sun at an enemy's ships would be a very interesting demonstration of optical principles.”

― Terry Pratchett, Small Gods

Oh so this is like nat.dev?

Yes and no:

  1. SOTA functionality is often released without API (eg: ChatGPT Code Interpreter, Bing Image Creator, Bard Multimodal Input, Claude Multifile Upload). We insist on using webapps so that you have full access to all functionality on launch day. We also made light/dark mode for each app, just for fun (Cmd+Shift+L (Aug update: currently broken in the GodMode rewrite, will fix))
  2. This is a secondary browser that can be pulled up with a keyboard shortcut (Cmd+Shift+G, customizable). Feels a LOT faster than having it live in a browser window somewhere and is easy to pull up/dismiss during long generations.
  3. Supports no-API models like Perplexity and Poe, and local models like LLaMa and Vicuna (via OobaBooga).
  4. No paywall, build from source.
  5. Fancy new features like PromptCritic (AI assisted prompt improvement)

Supported LLM Providers

Provider (default in bold) Notes
ChatGPT Defaults to "GPT4.5"!
Claude 2 Excellent, long context, multi document, fast model.
Perplexity The login is finnicky - login to Google on any of the other chats, and then reload (cmd+R) - it'll auto login. Hopefully they make it more intuitive/reliable in future.
Bing Microsoft's best. It's not the same as GPT-4!. We could use help normalizing its styling.
Bard Google's best. Bard's updates are... flaky
Quora Poe Added in #118
You.com Chat Added in #142
HuggingChat Offers Llama2, OpenAssistant
Vercel Chat Added in #117
Local/GGML Models (via OobaBooga) Requires Local Setup, see oobabooga docs
Phind Developer focused chat (temporarily disabled)
OpenRouter Access GPT4, Claude, PaLM, and open source models
OpenAssistant Coming Soon — Submit a PR!
Claude 1 Requires Beta Access
... What Else? Submit a New Issue!

Features and Usage

  • Keyboard Shortcuts:

    • Use Cmd+Shift+G for quick open and Cmd+Enter to submit.
    • Customize these shortcuts (thanks @davej!):
      • Quick Open
        • image
      • Submit can be toggled to use Enter (faster for quick chat replies) vs Cmd+Enter (easier to enter multiline prompts)
    • Cmd+Shift+L to toggle light/dark mode (not customizable for now)
    • Remember you can customize further by building from source!

  • Pane Resizing and Rearranging:

    • Resize the panes by clicking and dragging.
    • Use Cmd+1/2/3 to pop out individual webviews
    • Use Cmd +/- to zoom in/out globally
    • open up the panel on the bottom right to reorder panes or reset them to default
    • Cmd p to pin/unpin the window Always on Top

  • Model Toggle:

    • Enable/disable providers by accessing the context menu. The choice is saved for future sessions.
    • Supported models: ChatGPT, Bing, Bard, Claude 1/2, and more (see Supported LLM Providers above)

  • Support for oobabooga/text-generation-webui:

    • Initial support for oobabooga/text-generation-webui has been added.
    • Users need to follow the process outlined in the text-generation-webui repository, including downloading models (e.g. LLaMa-13B-GGML).
    • Run the model on http://127.0.0.1:7860/ before running it inside of the smol menubar.
    • The UI only supports one kind of prompt template. Contributions are welcome to make the templating customizable (see the Oobabooga.js provider).

  • Starting New Conversations:

    • Use Cmd+R to start a new conversation with a simple window refresh.

  • Prompt Critic: Uses Llama 2 to improve your prompting when you want it!

video demo

Download and Setup

You can:

  • download the precompiled binaries for MacOS: https://github.com/smol-ai/GodMode/releases/latest (sometimes Apple marks these as untrusted/damaged, just open them up in Applications and right-click-open to run it).
  • Or run it from source (instructions below)

When you first run the app:

  1. log into your Google account (once you log into your google account for chatgpt, you'l also be logged in to Bard, Perplexity, Anthropic, etc).
  2. For Bing, after you log in to your Microsoft account, you'll need to refresh to get into the Bing Chat screen. It's a little finnicky at first try but it works.

seeking contributors!

please see https://github.com/smol-ai/GodMode/blob/main/CONTRIBUTING.md

build from source

If you want to build from source, you will need to clone the repo and open the project folder:

  1. Clone the repository and navigate to the project folder:

    git clone https://github.com/smol-ai/GodMode.git
cd menubar
npm install
# On Windows, you may also need Squirrel - these are old instructions, we would love a Windows volunteer to verify
# npm install electron-squirrel-startup

npm run start # to run in development, locally

  2. Generate binaries:

    npm run package # https://electron-react-boilerplate.js.org/docs/packaging
# ts-node scripts/clean.js dist clears the webpackPaths.distPath, webpackPaths.buildPath, webpackPaths.dllPath
# npm run build outputs to /release/app/dist/main
# electron-builder build --publish never builds and code signs the app.

# this is mostly for swyx to publish the official codesigned and notarized releases

    The outputs will be located in the /release/build directory.

windows/linux builds

by default we're mac only - i only have a mac sorry. (we are seeking a "Windows Maintainer"! and someone to help make this work on Arch Linux)

Please let us know if you hvave windows/linux instructions.

Related project

I only later heard about https://github.com/sunner/ChatALL which is cool but I think defaulting to a menbuar/webview experience is better - you get to use full features like Code Interpreter and Claude 2 file upload when they come out, without waiting for API