CognifyCL

Seamlessly build your public knowledge graph—without compromising on privacy.

---

Overview

CognifyCL is a privacy-conscious, developer-friendly system for turning browser activity into a structured, queryable archive. It captures visited URLs via a Chrome extension, persists them through a local sync bridge, and surfaces them in a Next.js web interface that supports on-device summarization using llama.cpp.

Rather than routing your data through third-party APIs or cloud services, CognifyCL runs everything locally: data collection, storage, visualization, and LLM inference. A fixed category graph provides intuitive structure, while a split-pane layout enables seamless reading and summarization workflows. It's ideal for personal knowledge management, research traceability, or anyone looking to build a local-first toolchain for information curation.

---

Key Capabilities

• Local-Only Architecture: No cloud services. All processing stays on your machine.
• Automatic Link Capture: Chrome extension records every page you visit.
• Fixed Category Graph: Links are displayed in a predefined graph of topics.
• Split-View Interface: Read content and view summaries side by side.
• On-Demand LLM Summarization: Summaries are generated using a local TinyLlama model.

---

System Architecture

• Chrome Extension Captures visited URLs and sends them to a local sync server.

• Sync Bridge (Node.js/Express) Accepts and stores entries (url, title, summary, category) in entries.json.

• Web Application (Next.js) Displays links in a fixed graph layout with an interactive UI for summaries.

• Summarization Engine Uses llama.cpp with TinyLlama to run local LLM inference.

---

Demo Link :

---

Prerequisites

• Node.js
• A compiled version of llama.cpp
• Chrome browser (for the extension)

---

Installation and Setup

1. Clone the Repository

git clone https://github.com/PranavPipariya/CognifyCL.git
cd CognifyCL

---

2. Install Dependencies

Web Application

cd webapp
npm install

Sync Bridge

cd ../sync-bridge
npm install

---

3. Configure the Summarization Engine

Inside CognifyCL run:

mkdir llama-engine
cd llama-engine

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp

once you're in /CognifyCL/llama-engine/llama.cpp, then run:

mkdir build
cd build
cmake .. -DLLAMA_METAL=ON
cmake --build . --config Release

then again go to llama.cpp directory i.e /CognifyCL/llama-engine/llama.cpp by doing:

cd ..

then inside llama.cpp run:

mkdir -p models/gemma
cd models/gemma

place the model you downloaded in gemma. do not forget to rename it as tinyllama-chat.gguf !

thats it now your summarisation feature will work.

• Download the TinyLlama model:

  TinyLlama 1.1B Q2_K (GGUF) — Hugging Face

• Move the downloaded file to:

  llama-engine/llama.cpp/models/gemma/tinyllama-chat.gguf
  

---

4. Start the Services

Sync Bridge

cd sync-bridge
npm run start

Runs at: http://localhost:3001

Web Application

cd ../webapp
npm run dev

Runs at: http://localhost:3000

---

5. Load the Chrome Extension

1. Open chrome://extensions/
2. Enable Developer Mode
3. Click Load unpacked
4. Select the extension/ directory

---

Usage

1. As you browse, the extension captures each link.

2. Visit http://localhost:3000/dashboard to explore your history.

3. The dashboard presents a fixed category graph layout.

4. Click a category to view links under it.

5. Click a link to open a split view:

   • Right: the full web page
   • Left: a panel of category cards

6. Click a card to trigger a local LLM summary via TinyLlama.

---

Common Issues

• LLM not responding:

  • Verify the model is named tinyllama-chat.gguf and located in: llama-engine/llama.cpp/models/gemma/tinyllama-chat.gguf

---

Acknowledgements

• llama.cpp — lightweight, high-performance local inference engine
• Next.js — full-stack React framework for the frontend
• TheBloke — provider of quantized TinyLlama GGUF models

---