DalaiLLMA - LLM Conversation Sentiment Dashboard

A personal analytics dashboard that visualizes sentiment, wellbeing, and patterns from your Claude and ChatGPT conversation history.

Features

• Sentiment Tracking: Monitor emotional tone over time (hopeful vs despair language)
• Wellbeing Score: Composite metric combining sentiment, agency, and other indicators
• Word Clouds: Visual representation of frequent topics per month
• People Analysis: Track who you discuss and associated sentiment
• Category Breakdown: See what topics dominate your conversations (work, relationships, mental health, etc.)
• Events Timeline: Significant life events extracted from conversations
• Drama Triangle: Karpman triangle analysis (victim/persecutor/rescuer patterns)
• Victim vs Empowered Language: Track agency in your communication patterns
• LLM Insights: AI-generated deeper analysis using Claude API (optional, parallel processing)

Quick Start

1. Clone and Install

git clone https://github.com/kryptokommunist/dalaillma
cd dalaillma
npm install
npm run setup   # Creates data/ and output/ directories

2. Export Your Conversation Data

Claude.ai:

• Go to Settings → Export data
• Download and extract the zip
• Copy conversations.json to data/anthropic/

ChatGPT:

• Go to Settings → Data controls → Export data
• Wait for email, download the zip
• Copy the Conversations__*.zip file to data/openai/ (no need to extract)

3. Process Data (Basic Analysis - No API Required)

npm run process    # Analyzes conversations
npm run build      # Generates dashboard HTML

Or run both in one command:

npm run all

This generates:

• output/dashboard_data.json - Basic sentiment, word frequencies, events
• dashboard.html - Your personalized dashboard

4. View Your Dashboard

npm run serve
# Open http://localhost:3000/dashboard.html

Or just open dashboard.html directly in your browser (works with file:// protocol).

5. LLM Analysis (Optional - Requires Anthropic API)

For deeper AI-generated insights:

# Set your Anthropic API credentials
export ANTHROPIC_BASE_URL="https://api.anthropic.com/"
export ANTHROPIC_AUTH_TOKEN="your-api-key"
export ANTHROPIC_MODEL="claude-3-sonnet-20240229"

# Run full analysis (processes in parallel)
npm run full

This generates additional files:

• output/category_analysis.json - LLM-categorized messages per month
• output/llm_insights.json - Overall patterns and recommendations
• output/person_insights.json - Relationship analysis

Processing time: ~5 minutes for 30 months with parallel processing

Directory Structure

dalaillma/
├── data/                    # Your exports (gitignored)
│   ├── anthropic/
│   │   └── conversations.json
│   └── openai/
│       └── Conversations__*.zip
├── output/                  # Generated data (gitignored)
│   ├── dashboard_data.json
│   ├── category_analysis.json    # (optional, from LLM analysis)
│   ├── llm_insights.json         # (optional, from LLM analysis)
│   └── person_insights.json      # (optional, from LLM analysis)
├── scripts/
│   ├── process_data.js      # Basic analysis (no API)
│   ├── analyze_all.js       # Parallel LLM analysis
│   └── build_dashboard.js   # Generate HTML from data
├── templates/
│   └── dashboard.html       # Dashboard template
├── dashboard.html           # Generated output (gitignored)
├── dalaillma-logo.png
├── README.md
└── package.json

NPM Scripts

Script	Description
npm run setup	Create data directories
npm run process	Basic sentiment analysis (no API)
npm run analyze	LLM analysis (requires API)
npm run analyze:force	Re-analyze all months
npm run build	Generate dashboard HTML
npm run all	Process + build (no API)
npm run full	Process + analyze + build (with API)
npm run serve	Start local server

Environment Variables

# Optional: Custom data/output paths
export DATA_DIR="./data"
export OUTPUT_DIR="./output"

# Required for LLM analysis only
export ANTHROPIC_BASE_URL="https://api.anthropic.com/"
export ANTHROPIC_AUTH_TOKEN="your-api-key"
export ANTHROPIC_MODEL="claude-3-sonnet-20240229"

How It Works

Basic Analysis (process_data.js)

• Parses Claude and ChatGPT conversation exports
• Calculates sentiment using keyword matching (hopeful vs despair words)
• Tracks agency via phrase patterns ("I can" vs "I have to")
• Extracts significant events from conversation titles
• Counts people mentions and associated sentiment
• Generates word frequency data

LLM Analysis (analyze_all.js)

• Uses Claude API to categorize messages (relationships, work, mental health, etc.)
• Analyzes victim vs empowered language patterns
• Generates monthly insights and themes
• Creates relationship summaries for mentioned people
• Produces overall patterns and recommendations

Dashboard Builder (build_dashboard.js)

• Reads processed data from output/
• Injects data into the HTML template
• Creates a self-contained dashboard.html file
• Works with file:// protocol (no server required)

Data Privacy

• All processing happens locally - no data sent anywhere except Anthropic API (if using LLM analysis)
• Your data stays in data/ which is gitignored
• Generated outputs in output/ are gitignored
• The generated dashboard.html is gitignored
• Never commit API keys, conversation exports, or personal insights

Testing

Verify dashboard loads correctly with Playwright:

npm install playwright
npm test

License

MIT