🧠 Personal Knowledge Base (PKB) Links Automation

This project is a Personal Knowledge Base (PKB) Links tool that automatically collects, summarizes, tags, and organizes links into category-specific Markdown files for easy reference and search.

---

🚀 Features

• Fetch titles of links from new_links.md
• Extract article/repo/blog content with robust handling for:
  • Medium & Blogs (newspaper3k)
  • GitHub repos (README extraction)
  • LinkedIn (skipped with warning)
  • Generic websites (paragraph scraping)
• Summarize content using a local Llama model with configurable summary length:
  • short (~1–2 sentences)
  • medium (~2–3 sentences)
  • long (~3–5 sentences)
• Categorize links based on keyword → category + tags mapping
• Store categorized links into categories/*.md with title, summary, tags, and timestamp
• Full-text search across titles, summaries, and tags
• Logs all activity to logs/pkb_log.json

---

📂 Project Structure

• category_keywords.py → Defines categories and keyword-to-tag mapping
• pkb_main.py → Main script: process links, categorize, store, search
• summarizer.py → Extracts & summarizes content with error handling and logging
• requirements.txt → Python dependencies

---

🛠️ Workflow

High-Level Workflow

Detailed Workflow (Content Extraction & Summarization)

---

⚙️ Installation

git clone <repo-url>
cd personal-knowledge-base

# Create virtual environment (recommended)
python -m venv venv
source venv/bin/activate   # On Linux/Mac
venv\Scripts\activate      # On Windows

# Install dependencies
pip install -r requirements.txt


You will also need a **local Llama model server** running with `llama.cpp` exposed at:  

http://localhost:12434/engines/llama.cpp/v1/chat/completions

---

## ▶️ Usage

1. Add new links to `links/new_links.md`
2. Run the script:
   ```bash
   python pkb_main.py --process --summary-length short

3. Categorized links with summaries will be saved into categories/
4. Search stored links interactively by entering queries

---

📌 Example

Input (links/new_links.md):

https://github.com/oumi-ai/oumi
https://markharrison.io/blog/model-context-protocol

Output (categories/AI.md):

# AI

- [Oumi: Lightweight memory layer](https://github.com/oumi-ai/oumi) : Simple agent memory layer for LLMs - Added on: [2025-09-21]
- [Model Context Protocol](https://markharrison.io/blog/model-context-protocol) : Standard for LLM interoperability and context sharing - Added on: [2025-09-21]

---

🔍 Search

You can search links across categories:

python pkb_main.py --search "agent"

Output:

Oumi: Lightweight memory layer (AI): https://github.com/oumi-ai/oumi
Model Context Protocol (AI): https://markharrison.io/blog/model-context-protocol

---

📝 Notes

• LinkedIn scraping is skipped (manual input required).
• Summarization requires a locally running Docker Model Runner (DMR) Llama model.
• Logs are stored in logs/pkb_log.json.
• Search works across titles, summaries, and tags.
• This project only to prototype only implementation with help of AI Assitance.

---

📖 License