Skip to content

Latest commit

 

History

History
56 lines (37 loc) · 2.33 KB

README.md

File metadata and controls

56 lines (37 loc) · 2.33 KB

Supermind

An open source autonomous agent you can customize entirely from text files -- no coding required. The framework is light, simple and modular, so you can extend the code as you like.

2021-11-29_02-45-46.mp4

Getting Started

Clone the repo

git clone https://github.com/DaVinciDreams/Thales

Then, grab an API key from Open AI. https://beta.openai.com/signup

Copy and paste the .env.default file, rename to .env and put your API key in it.

Now install dependencies and run the project

npm install
npm start // start with default agent

Contibuting

Contributions welcome

FOR PROGRAMMERS

Please look below to TODOs, and contribute what you're interested in. If you have other ideas for features, please give them a try and submit as a PR. Together we can make something really incredible!

FOR NON PROGRAMMERS

You can create your own agent entirely by modifying the text prompts, without needing more than to change a few lines of configuration. First, copy the folder in agents directory. Then add a command in the package.json -- you can copy one of the existing commands, and change the AGENT environment variable to the name of your personality. Modify the text files -- you can do this while the agent is running -- and explore the responses and history to tune your agent.

Here's a great place to start on your prompt engineering journey: https://www.gwern.net/GPT-3

STATES AND STATE MACHINES

TODO

  • Interface for adding and editing agents from UI

  • Add new agents through discord / command

  • Switch agent through discord / command

  • handle non-sequitors

  • Relationship matrix (including gradients)

  • Block repeated inappropriate use + build up enemy

  • Identify if question is expert knowledge or not

    • Neural search for deep knowledge -- if expert knowledge, respond with "good question", "let me think about that..." and do knowledge search

  • handle long user input, shorten if it's too long

  • Add up the length of all of the txt files to make sure they aren't super long in debug mode, throw warning if too long or force smaller context / remove memory

  • Set creativity, presence and frequency for each prompt type

  • Storage adapter to pull all data from either