Distraction-free YouTube study workspace with local AI note indexing and progressive concept mapping.
Reloop turns YouTube video lectures into an active, distraction-free study station. Its landing page at / accepts a YouTube link and embeds the actual study workspace at /focus as a live demo. The workspace extracts YouTube captions, summarizes 15-second lecture segments with local models (WebLLM via WebGPU or local Ollama), and progressively generates a multi-branch 2D concept map using Mermaid.js.
- Local-First AI Execution: Runs in-browser via WebLLM (
SmolLM2-360MorLlama-3.2-1B) utilizing WebGPU, or connects to a local Ollama instance (localhost:11434). Zero cloud API keys required. - Progressive Concept Mapping: Generates a 2D hierarchical flowchart (
flowchart TD) in real-time, grouping every 45s of lecture into topic phases with core concept hexagons and mechanism/insight sub-branches. - Automated Lecture Notes: Chronologically indexes bullet-point notes for each segment as you watch.
- Focus & Fullscreen Guard: Pauses video and displays an active lock modal if fullscreen is exited or the browser tab loses focus.
- Unlimited IndexedDB Storage: Uses
indexedDBwith persistent storage (ReloopBranchingDB_v4) to cache transcripts, notes, and concept graphs with quick 1-click resume. - Figma-Style Canvas: Unrestricted full-size SVG rendering with 10% to 800% zoom, mouse-wheel / trackpad pinch zoom, and click-and-drag panning.
- Minimalist Light Design: Clean editorial interface inspired by Linear, Notion, and Figma, built with Inter and JetBrains Mono on an architectural dot-grid canvas.
Reloop/
├── server.js # Express backend (caption extraction + Ollama proxy)
├── package.json # Dependencies and scripts
├── .gitignore # Leakage prevention & dependency exclusion
├── README.md # Project documentation
└── public/
├── landing.html # Public landing page
├── landing.css # Landing page styling
├── landing.js # Landing page interactions
├── youtube.js # Shared YouTube URL parsing and normalization
├── index.html # Focus-session workspace (served at /focus)
├── style.css # Minimalist Light Theme styles
└── app.js # Client logic (WebLLM, YouTube player, Mermaid canvas)
Reloop supports two local AI options:
WebLLM runs 100% inside your browser using WebGPU. You do not need to download or install any external model tools via terminal.
- Open Reloop in a WebGPU-enabled browser (Chrome, Edge, or Brave).
- Select one of the in-browser engines from the dropdown:
Local Engine: SmolLM2 360M(Fastest, ~360 MB, recommended for instant response)Local Engine: Llama 3.2 1B(~880 MB, deeper contextual reasoning)
- When you start your study session, the browser automatically downloads the model weights and caches them locally in your browser's IndexedDB. Subsequent runs load instantly from the cache.
If you prefer using Ollama on your system (localhost:11434):
-
Install Ollama:
- macOS / Linux:
curl -fsSL https://ollama.com/install.sh | sh - Windows: Download from ollama.com.
- macOS / Linux:
-
Pull the model: Download the default model used by Reloop:
ollama pull llama3.2
Optional smaller/alternative models:
# Lightweight 1B version ollama pull llama3.2:1b # Ultra-lightweight SmolLM2 ollama pull smollm2:360m
-
Start the Ollama server:
ollama serve
(If you use the Ollama Desktop app, it starts automatically).
-
In the Reloop topbar, select
Localhost: Ollama (11434)from the engine dropdown. Reloop will automatically detect your pulled model.
- Node.js (v18 or higher recommended)
- A modern browser with WebGPU support (Chrome, Edge, Brave, or Chromium-based)
- (Optional) Ollama installed if using external model inference
-
Clone the repository:
git clone https://github.com/CodeByRachit/Reloop.git cd Reloop -
Install dependencies:
npm install
-
Start the application:
npm start
-
Open your browser and navigate to the landing page:
http://localhost:3000
- Select your preferred local engine from the toolbar (
SmolLM2 360M,Llama 3.2 1B, orLocalhost: Ollama). - Paste any YouTube lecture URL that has captions/subtitles enabled.
- Click Enter Focus Session to launch fullscreen mode.
- Watch the lecture—Reloop will automatically build timestamped notes and grow the concept map step-by-step.
- Use the canvas controls (
+,-,100%) or scroll wheel / click-and-drag to pan and zoom from 10% to 800%.
MIT