Neurite

Introduction

💡 Neurite unleashes a new dimension of digital interface...

...the fractal dimension.

Available through Github Pages: Neurite

⚠️ Warning: Contains zooming, flashing lights, and colors which may not currently be suitable for people with photosensitive epilepsy.

See the User Guide for instructions on running locally

🌱 This is an open-source project in early alpha, we are looking to grow our team! 🚧

feel free to visit our Discord!

Table of Contents

1. Introduction
2. How to Use Neurite
3. FractalGPT
4. Local Ai
5. User Guide
6. Local Server Setup
7. Future of Neurite
8. Contact

How to Use Neurite

📚 Neurite lets you embed anything - text, photos, audio, video, PDFs, or i-frame content directly into the Mandelbrot set.

Drag and drop local files or web content straight into the fractal

Supports embedded links. Ctrl+V to paste an embed link

Controls

*Full controls are listed in the ? tab.

Shift + double click to create nodes.

Shift to freeze nodes.

Double click to anchor node.

Hold Shift + click two nodes to connect.

Alt + double click to create an Ai node.

FractalGPT

Not your typical chat interface.

• Web of thought reasoning.
• Non-linear, rhizomatic memory.
• Limitless Ai Agents.
• Fractal Interface.
• Custom tool creation.

Ai that doesn't get lost.

Neurite combines a number of prompting techniques that allow for finer control over the focus of your Ai conversation.

To interact with the Ai, you either need an OpenAi API key, or you can install an Ai directly in browser!

• The LLM’s responses format themselves into notes and connections within the fractal mind map. This emergent graph structure represents the AI’s memory.

  (Long term memory for conversation with Ai)

• Breaking up the AI's output into a chain of thought reasoning correlates to improved Ai response.

• Mind mapping combined with our fractal interface goes beyond chain of thought reasoning to produce an interactive fractal web of thought.

• Notes are easily returned to wherever they are at any point in the fractal via our search bar, or by clicking on their syntax highlighted titles in the notes tab.

• The Mandelbrot set acts as the terrain for your mind-map to grow into an interconnected tapestry of thoughts, references, and creations!

Neurite supports the following Ai features

1. Auto Mode
2. Node based memory.
3. HTML/JS and In-browser Python (pyodide) environment for directly rendering GPT's code output.
4. Web Search (requires Google programmable search api key and search engine id)

(to search without an API key, send a url within the prompt input or past an embed link and the webpage will display without going through Google.)

5. Webpage and PDF text extraction. (see the localhost servers documentation for how to set up)
6. Wikipedia Summaries (see the localhost servers)
7. Wolfram Alpha Results. (see the localhost servers)

All API keys can be input through the Ai tab of the menu dropdown.

Local Ai

Above is an example of conversation history being shared between GPT-4 and a Local Ai model, Red Pajamas, run through webLLM.

Ai Nodes allow for the creation of an arbitrary number of Ai Agents with the ability to connect their memories together.

• Alt + Double click creates a node with a more traditional AI chat interface. (you can create multiple)
• The AI node will retain context from any other text or AI nodes that connect to it. (Hold Shift + Click two node windows to connect them together)
• This means the ability to connect multiple chat interfaces together, and the ability to retain textual memories without having to re-copy and re-paste them every time.
• Supports webLLM and OpenAi models concurrently.

Neurite currently supports OpenAi, Red Pajamas 3B, and Vicuna 7B.

• webLLM runs Ai models directly in browser on the GPU or CPU.
• We are also running an embeddings model locally through Transformers.js for local webpage/pdf extraction.
• This is an early implementation.

See webLLM's documentation for how to run the models on your GPU.

Above is an example of combining the Wolfram Alpha API with GPT-4.

Wolfram, Wikipedia, and Webpage/PDF vector database plugins are accessed through the localhost servers.

User Guide

To run Neurite locally,

either:

1. Clone the repository
2. Open the index.html file in your browser (note: local AI features will not be enabled)

Or, to enable local AI features without using our Github pages host above:

1. Navigate to the cloned directory in your terminal (either main or local-ai branch)
2. Install any necessary global dependencies when running the below,

npm install
npm run build
npm run start

Local Server Setup

Download Localhost Servers here.

The servers enable plugin integration for webpage/pdf text extraction, Wolfram Alpha, and Wikipedia.

• The Wikipedia, Wolfram, and web scrape servers require Python and Node.js
• The readme in the Locahost Servers folder explains how to run all of the servers at once.
• Without any servers running, Wolfram, Wiki, and Webpage extractions will not function.
• All servers can be run with one command. Download the Localhost Servers folder, or find it in your cloned directory. Once navigated to the Localhost Servers folder in your command line, run

  node start_servers.js

This will run all of the servers at once. They'll work on our Github pages host of Neurite, or any other way you locally access the interface.

In the '?' tab, the AI HOW-TO checkbox will send a context message to the AI that allows it to answer questions about Neurite.

Notes

• Search: Enter a search query to locate windows containing that text. The search results display to the left of the menu. Click on a search result to zoom your view to where the corresponding window is positioned within the fractal.
• Saving: Saving is currently a work-in-progress. Your networks will can be saved in the browser cache, or downloaded as a .txt file. Drag the .txt file into the save box, click load on the imported file in the save box, then click the second load button to display the network. Non-textual content currently needs to be re-inserted.
• Zettelkasten: The Zettelkasten method enables nodes to be created by typing into the main text area. The AI follows this format to create and connect its responses together.

The Zettelkasten note taking method accesses tapestry of thought reasoning for Ai, all within our real time Mandelbrot set visualization.

• This diagram represents how the AI's response incorperates note-taking to retain a non-linear conversation history.
• We feed both the recent conversation as well as any matched notes that are not already in the recent conversation as part of the context message to the Ai.
• This means retaining relevant information from any part of the conversation regardless of the Ai's context window size.
• Still, advancements in context window size only further enable Neurite's performance.

Future of Neurite

Neurite is a recursive environment for generating ideas. The project combines fractals, mind-mapping, and ai to enable an experimental, yet powerful playground for research, code generation, writing, learning, and visualizing thought.

🚧 Neurite is in active development. 🚧

✅ webLLM

• webLLM localhost

✅ local embeddings

• VR
• deeper fractal zoom
• Ai cursor control

✅ color selection

✅ adjust width, length, max lines, speed

✅ screen recording

• accessibility
• auto embed formats
• custom equations for the fractal
• drawing-tool

🔄 user feedback/bug fixes

🚧

Contact

If you are a developer who is interested in contributing to this project, contact us at [email protected] or visit our discord 🔗

full video here