Vision โข Core Features โข Architecture โข Tech Stack โข Getting Started โข Privacy Model
The modern web is built for engagement and stimulation, but for neurodivergent individualsโespecially those with ADHD, Autism, or Dyslexiaโthis digital environment frequently triggers severe cognitive overload, visual fatigue, and comprehension barriers.
Traditional accessibility extensions attempt to address this by transmitting reading data to third-party cloud APIs. This introduces latency, breaks offline functionality, and fundamentally compromises user privacy.
NeuroEdge AI solves this natively at the edge. By combining local LLM inference engines (phi3:mini via Ollama) with GPU-accelerated Document Object Model (DOM) transformations, the extension restructures typographical physics, visual layouts, and linguistic complexity entirely on the user's local machine.
- Zero Cloud Dependencies: All processing occurs locally on your hardware.
- Zero Telemetry: Reading history and highlighted content never leave
localhost. - True Accessibility: Real-time visual and linguistic transformation tailored to cognitive needs.
- Highlight dense corporate jargon, convoluted legalese, or complex academic literature.
- Content is routed locally to an isolated Python FastAPI backend running a quantized local model (
phi3:mini). - Synthesizes the text into clean, digestible language and performs seamless inline replacement in milliseconds without reloading the page.
- Bypasses strict Chromium Content Security Policies (CSPs) by dynamically injecting native
OpenDyslexicfont binaries into the browser runtime via theFontFaceAPI. - Restyles active web typography with bottom-weighted characters to anchor visual saccades and prevent letter inversions or word flipping.
- An optimized algorithmic
TreeWalkertraverses textual DOM elements in real-time. - Fixates reading gaze by applying high-contrast
font-weight: 900styling to word prefixes while attenuating trailing suffixes. - Guides the visual cortex across sentences to preserve attention and reading speed without breaking host layout constraints.
- Scope Selector:
- Global Mode: Automatically applies your preferred reading adaptations to every newly opened tab.
- Local Mode: Applies overrides strictly to the active viewport, preserving original site styles elsewhere.
- Cognitive Continuance: Background listeners track scroll position and synthesized text states. If a tab closes accidentally, re-opening the URL restores the tailored reading state instantly.
flowchart TD
subgraph Browser["๐ฅ๏ธ Chromium Browser Environment (Chrome / Edge)"]
User(["๐ค User"])
Webpage["๐ Active Webpage (DOM)"]
subgraph Extension["๐งฉ Neuro-Assist Extension (Plasmo + React + TS)"]
Content["Content Script\n(TreeWalker & DOM Parser)"]
Engine["Accessibility Engine\n(OpenDyslexic & Bionic Styler)"]
Storage[("Local Storage\n@plasmohq/storage")]
end
end
subgraph Edge["๐ Local Edge AI Environment (Offline / Zero-Telemetry)"]
API["โก FastAPI Server\n(localhost:8000)"]
Ollama["๐ง Ollama Engine\n(phi3:mini)"]
end
User -->|"Highlights text & toggles modes"| Extension
Extension <-->|"Persist scopes & states"| Storage
Engine -->|"Injects fonts & bionic styling"| Webpage
Content -->|"1. POST /simplify (Local JSON)"| API
API -->|"2. Prompt evaluation"| Ollama
Ollama -->|"3. Synthesized plain text"| API
API -->|"4. Return response"| Content
Content -->|"5. Inline DOM replacement"| Webpage
Webpage -->|"Frictionless, accessible reading"| User
classDef browser fill:#0f172a,stroke:#38bdf8,stroke-width:1.5px,color:#f8fafc;
classDef ext fill:#1e1b4b,stroke:#818cf8,stroke-width:1.5px,color:#f8fafc;
classDef edge fill:#064e3b,stroke:#34d399,stroke-width:1.5px,color:#f8fafc;
classDef component fill:#1e293b,stroke:#64748b,stroke-width:1px,color:#f8fafc;
class Browser browser;
class Extension ext;
class Edge edge;
class Content,Engine,Storage,API,Ollama,Webpage component;
| Layer | Technology | Purpose |
|---|---|---|
| Extension Framework | Plasmo (React 18, TypeScript) | Modular MV3 extension framework with hot reload |
| UI & Styling | Tailwind CSS (Glassmorphism Dark Theme) | Accessible, modern control popup and overlay UI |
| State Persistence | @plasmohq/storage |
Cross-tab synchronized session and scope memory |
| Local AI Gateway | FastAPI + Uvicorn (Python 3) | Lightweight, asynchronous local HTTP inference bridge |
| Local LLM Runner | Ollama (phi3:mini) |
Hardware-accelerated on-device language model inference |
| DOM Manipulation | Native JavaScript TreeWalker & FontFace API |
Zero-dependency, performant DOM parsing and font rendering |
-
Navigate to the backend directory:
cd edge-backend -
Create and activate a virtual environment:
# On macOS/Linux: python3 -m venv venv source venv/bin/activate # On Windows: python -m venv venv venv\Scripts\activate
-
Install Python dependencies:
pip install fastapi uvicorn requests
-
Pull the target language model via Ollama:
ollama pull phi3:mini
-
Launch the local Edge API:
python main.py
The API server runs at
http://localhost:8000with interactive docs athttp://localhost:8000/docs.
-
Navigate to the extension directory:
cd edge-assistant -
Install project dependencies:
npm install
-
Build the production bundle:
npm run build
For development with live reloading, run
npm run devinstead.
- Open Google Chrome, Microsoft Edge, or Brave and navigate to
chrome://extensions/. - Toggle on Developer mode in the top-right corner.
- Click Load unpacked.
- Select the build output directory:
(If using
edge-assistant/build/chrome-mv3-prodnpm run dev, selectedge-assistant/build/chrome-mv3-dev).
- Pin the Extension: Click the puzzle icon in your browser toolbar and pin Neuro-Assist.
- Configure Reading Modes:
- Click the extension icon to open the Glassmorphism Control Center.
- Toggle OpenDyslexic Font to inject dyslexia-friendly typography.
- Toggle Bionic Focus to activate fixation prefix highlighting.
- Select your Scope (
Globalacross all tabs orLocalfor the current page).
- Simplify Complex Text:
- Highlight any challenging passage or jargon-heavy paragraph.
- Click the inline simplify trigger to receive an instant, plain-language breakdown generated locally by
phi3:mini.
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ ON-DEVICE PRIVACY BOUNDARY โ
โ โ
โ Webpage Content โโโบ Browser Extension โโโบ Localhost:8000 โ
โ โ โ
โ โผ โ
โ Inline Replacement โโโ Transformed Text โโโ Ollama Engine โ
โ โ
โ [ NO CLOUD โข NO TELEMETRY โข OFFLINE ] โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
- Strict Network Containment: All network requests are routed exclusively to
localhost:8000. No remote analytics, tracking pixels, or external API calls exist within the codebase. - Local DOM Modification: Text substitutions and font injections happen in-memory within your active browser tab.
- Absolute Data Sovereignty: Your reading habits, highlighted materials, and personal data never leave your physical device.
This project is licensed under the MIT License.