Imagine a world where multiple Large Language Models (LLMs) collaborate together to help us humans with various tasks.
Well, this application is a step towards that world.
- Powered by the OpenAI GPT-3 API
- Current members:
@davinci
,@dalle
, and@codex
- Assistants are given their own distinct personalities, behaviors, and capabilities through dynamically-created generation prompts.
- Subject to tweaks and changes over time, so read through the following files instead: ./src/util/chat/users
- Tip: if you want to see the final prompt text that was created for each assistant message, I am printing them as 'error' to the JS console, lol
@coordinator
is a special assistant who dictates who should respond to the user's message by tagging other assistants- Tip: Directly tag an assistant (ex:
@davinci
) to only have them respond (this skips the coordinator)
@gen_image
is an assistant helper that directly uses the OpenAI DALL·E 2 API, which can generate images from text prompts.@dalle
(the chatting assistant) can help you generate these text prompts to generate images as you'd like.- Tip: you can also directly prompt the OpenAI DALL·E 2 API just like
@dalle
did (<gen_image>A picture of a cat.<gen_image>
)
- Hover over a message to visualize the messages that were used to provide context
- Coordinator messages are ignored by default - they won't be used as generation context for other assistants because that would waste tokens
- You can manually ignore messages by clicking the 'eye' icon at the bottom-right of a message
- The UI for the chat message bubbles supports Markdown formatting
Install Node.JS (version ^18 required) and Yarn:
nvm install 18
# Optional: set this version as default:
nvm alias default node
Install Yarn:
npm install -g yarn
Install the rest of the project dependencies
yarn install
Important!! - You will have to put your OpenAI API Key inside the .env
file.
- If forking this repository, run
git update-index --assume-unchanged .env
to make sure you don't accidentally push your API key to the repo
- hot-code reloading, error reporting, etc.
yarn start
yarn build
yarn lint
yarn format
You may use this software and/or modify its functionality as you wish, but I ask you to keep the in-app credits intact and include a link to this repository in any derivative work.