🔮 Claude Chat Knowledge Extractor

Transform your Claude chat history into reusable AI skills, templates, and knowledge bases.

This project provides a complete pipeline to export, clean, classify, and package your Claude (Anthropic) chat history into structured knowledge that can be used to create AI skills, templates, and best practices documentation.

🎥 Video Tutorial

Watch the full walkthrough on YouTube:

🇻🇳 Vietnamese tutorial — Covers export, pipeline setup, and skill packaging.

📖 What This Does

Step	Script	Output	Description
1️⃣ Export	Manual	fie1/, file2/	Raw Claude data export (JSON)
2️⃣ Clean	step1_clean_sessions.py	output/	Split into individual session files by month
3️⃣ Stats	step2_generate_stats.py	reports/	Generate statistical reports (MD + HTML dashboard)
4️⃣ Extract	step3_extract_knowledge.py	output_topic/	Filter & classify by your keywords
5️⃣ Analyze	step4_analyze_patterns.py	analysis_topic/	Deep analysis → patterns & templates

🚀 Quick Start

Prerequisites

Python 3.10+
# No external dependencies needed — uses only standard library

Step 1: Export Your Claude Data

1. Go to claude.ai/settings
2. Click Export Data
3. Download the zip and extract to this directory:

   export-claude/
   ├── fie1/              # Personal chat history
   │   └── conversations.json
   └── file2/             # Project-related data
       ├── conversations.json
       ├── memories.json
       ├── users.json
       └── projects/
   

Step 2: Clean & Split Sessions

python scripts/step1_clean_sessions.py

Output: output/ directory with sessions organized by year-month:

output/
├── personal/2025-06/conv_xxx.json
├── personal/2025-07/conv_xxx.json
├── project_chats/2024-04/chat_xxx.json
└── ...

Each session file contains:

• meta: uuid, name, date, topic hints, token estimate
• turns: list of chat turns (role + text)
• full_text: complete conversation in plain text

Step 3: Generate Statistics Report

python scripts/step2_generate_stats.py

Output: reports/

• statistics_report.md — Markdown report with tables
• statistics_dashboard.html — Interactive HTML dashboard with charts

Step 4: Extract Topic-Specific Knowledge

Edit step3_extract_knowledge.py to customize for your domain, then run:

python scripts/step3_extract_knowledge.py

Output: output_topic/

output_topic/
├── index.md                    # Master index
├── topic_setup.md              # Setup & configuration sessions
├── topic_api.md                # API-related sessions
└── topic_general.md            # Other sessions

Step 5: Analyze & Extract Patterns

Edit step4_analyze_patterns.py to customize pattern detection, then run:

python scripts/step4_analyze_patterns.py

Output: analysis_topic/

• pattern_analysis.md — Extracted patterns with examples
• patterns_data.json — Structured pattern data

🔧 Customization

Extracting Your Topic

Edit scripts/step3_extract_knowledge.py:

# Change these for your domain
TOPIC_KEYWORDS = [
    "your_product", "your_product_api",
    "your_sdk", "your_webhook",
]

SUBTOPICS = {
    "topic_setup": {
        "name": "Your Product - Setup",
        "keywords": ["setup", "install", "config"],
    },
    "topic_api": {
        "name": "Your Product - API",
        "keywords": ["api", "endpoint", "request"],
    },
    "topic_general": {
        "name": "Your Product - General",
        "keywords": [],  # Catch-all
    },
}

See scripts/TEMPLATE.md for full customization guide with examples (n8n, CRM, etc.).

Analyzing Your Patterns

Edit scripts/step4_analyze_patterns.py:

PATTERN_CATEGORIES = {
    "code_blocks": {
        "name": "Code Blocks",
        "regex": r"```(?:\w+)?\n(.*?)```",
        "flags": re.DOTALL,
    },
    "your_pattern": {
        "name": "Your Pattern Name",
        "regex": r"YOUR_REGEX_HERE",
    },
}

📁 Directory Structure

export-claude/
├── scripts/                      # Pipeline scripts (committed)
│   ├── step1_clean_sessions.py
│   ├── step2_generate_stats.py
│   ├── step3_extract_knowledge.py   # Customize keywords here
│   ├── step4_analyze_patterns.py    # Customize patterns here
│   └── TEMPLATE.md                  # Customization guide
├── fie1/                         # ⚠️ Raw personal export (gitignored)
├── file2/                        # ⚠️ Raw project export (gitignored)
├── output/                       # ⚠️ Cleaned sessions (gitignored)
├── output_topic/                 # ⚠️ Topic-extracted (gitignored)
├── reports/                      # ⚠️ Stats reports (gitignored)
├── analysis_topic/               # ⚠️ Analysis output (gitignored)
├── .gitignore                    # Excludes sensitive data
├── README.md                     # This file
└── AGENTS.md                     # AI agent instructions

🔒 Privacy & Security

All personal data is gitignored:

• Raw export files (fie1/, file2/)
• Processed sessions (output/)
• Extracted knowledge (output_topic/)
• Reports (reports/)

Only scripts and templates are committed.

Before sharing:

1. Verify .gitignore excludes all personal data
2. Run git status to confirm no sensitive files are staged
3. Consider anonymizing any examples in templates

📝 Use Cases

Personal Knowledge Management

• Extract your own expertise from chat history
• Create a searchable knowledge base
• Build AI skills that mirror your working style

Team Knowledge Sharing

• Aggregate team chat history
• Identify common patterns and solutions
• Create standardized templates and best practices

AI Skill Development

• Extract real-world patterns from actual conversations
• Build domain-specific AI skills
• Create comprehensive reference documentation

📄 License

MIT License — use freely, modify, and share.

🏢 Built by Diginno

We help Vietnamese SMEs automate operations with LarkSuite, n8n, and custom data pipelines.

• 🌐 Website: https://diginno.net/services
• 📧 Contact: Reach out for automation consulting

---

Transform your AI conversations into actionable knowledge.