tonepilot 0.2.3

Emotional intelligence for text generation - detect emotional tones and generate contextually appropriate responses

TonePilot

🧠 TonePilot

Emotionally Intelligent Prompt & Response Engine for AI Chatbots

TonePilot helps your chatbot understand the user’s tone and personality—and respond like a human would.
Whether you're building a customer support bot, a mental health assistant, or a flirty AI companion, TonePilot makes it emotionally resonant and personality-aware.

✨ Free, Open Source, and Built for Multi-LLM Integration.

---

🚀 Features

• 🎯 Emotion Detection – Detect nuanced tones like anxious, playful, confused using transformer-based models
• 🧠 Emotionally Matched Responses – Generate emotionally resonant replies using Gemini or Hugging Face
• 🧩 Pluggable Architecture – Easily switch between LLMs or swap in your own tone classifiers
• 💡 Prompt-Only OR Full Response Mode – Get just the enhanced prompt, or the complete reply
• 🛠️ CLI + Python API – Use it in scripts, backends, or with your own chat interface

---

📦 Installation

Install from PyPI:

pip install tonepilot

📋 View on PyPI: https://pypi.org/project/tonepilot/

Quick Start

Basic Usage (Prompt Generation Only)

# Generate an emotionally-aware prompt without API keys
tonepilot "I'm feeling overwhelmed with work deadlines"

Output:

📝 Input: I'm feeling overwhelmed with work deadlines

🏷️  Detected tags:
  - stressed: 0.456
  - anxious: 0.234

⚖️  Response tags and weights:
  - calming_supporter: 0.445
  - practical_helper: 0.289

🔍 Final prompt:
Respond with calm reassurance and practical guidance. Help organize thoughts and provide actionable steps.

User: I'm feeling overwhelmed with work deadlines
Assistant: (Aim to respond in about 89 words)

Full Response Generation

# Generate complete responses (requires API key)
tonepilot "I'm excited about my new job!" --mode gemini --respond true

Environment Setup

For Response Generation (Optional)

If you want to generate actual responses (not just prompts), set up API credentials:

Option 1: Environment Variable

export GOOGLE_API_KEY=your_api_key_here

Option 2: .env File

echo "GOOGLE_API_KEY=your_api_key_here" > .env

Get your API key from: Google AI Studio

CLI Usage

Basic Commands

# Default: HuggingFace mode, prompt only
tonepilot "Your text here"

# Generate full response with Gemini
tonepilot "Your text here" --mode gemini --respond true

# Generate full response with HuggingFace
tonepilot "Your text here" --mode hf --respond true

# Different boolean formats accepted
tonepilot "Text" --respond yes
tonepilot "Text" --respond 1 
tonepilot "Text" --respond false

Available Options

• --mode {hf,gemini}: Choose the response generation model (default: hf)
• --respond {true,false,yes,no,1,0}: Generate response or just prompt (default: false)

Python API

from tonepilot.core.tonepilot import TonePilotEngine

# Initialize engine
engine = TonePilotEngine(mode='hf', respond=False)

# Process text
result = engine.run("I'm nervous about my presentation tomorrow")

print("Detected emotions:", result['input_tags'])
print("Response emotions:", result['response_tags'])
print("Generated prompt:", result['final_prompt'])

# For response generation (requires API key)
engine_with_response = TonePilotEngine(mode='gemini', respond=True)
result = engine_with_response.run("I'm nervous about my presentation tomorrow")
print("Generated response:", result['response_text'])

Architecture

TonePilot uses a sophisticated multi-stage pipeline:

1. Emotion Detection: Zero-shot classification using BART-large-MNLI
2. Tone Mapping: BERT-based classifier maps input emotions to response personalities
3. Prompt Blending: Combines personality traits with weighted importance
4. Response Generation: Optional text generation using HuggingFace or Gemini models

Supported Emotions

Input Emotions: curious, angry, sad, excited, confused, hopeful, tired, scared, playful, assertive

Response Personalities: empathetic_listener, direct_ally, calming_supporter, practical_helper, and more

Model Downloads

TonePilot uses a custom-trained BERT classifier for tone mapping. The model is automatically downloaded on first use - no manual installation required!

BERT Tone Classifier

🚀 Quantized Model (Recommended)

• File: tonepilot_bert_classifier_quantized.pt (119 MB - 75% smaller!)
• Download: GitHub Releases
• Benefits: 4x faster deployment, 75% less memory usage, same accuracy

📋 Original Model

• File: tonepilot_bert_classifier.pt (475 MB)
• Download: GitHub Releases
• Purpose: Maps detected emotions to appropriate response personalities
• Training: Custom-trained on emotional response datasets

Automatic Model Management

TonePilot automatically handles model downloads and caching:

1. First Run: Downloads model to ~/.cache/tonepilot/ (one-time, ~475 MB)
2. Subsequent Runs: Uses cached model for instant loading
3. Fallback Locations: Also checks current directory and package directory

Manual Download (if needed):

# Download quantized model (recommended - 75% smaller)
wget https://github.com/sdurgi/tonepilot/releases/download/v0.1.0/tonepilot_bert_classifier_quantized.pt -P ~/.cache/tonepilot/

# Or place in current directory
curl -L -o tonepilot_bert_classifier_quantized.pt https://github.com/sdurgi/tonepilot/releases/download/v0.1.0/tonepilot_bert_classifier_quantized.pt

# Original model (if you need full precision)
curl -L -o tonepilot_bert_classifier.pt https://github.com/sdurgi/tonepilot/releases/download/v0.1.0/tonepilot_bert_classifier.pt

Note: Internet connection required only on first use for model download. TonePilot automatically downloads the optimized quantized model (119MB) for faster performance.

Examples

Different Emotional Contexts

# Sadness → Empathetic support
tonepilot "My dog passed away yesterday"

# Excitement → Enthusiastic encouragement  
tonepilot "I just got accepted to my dream university!"

# Confusion → Clear guidance
tonepilot "I don't understand this math problem at all"

# Anger → Calming and validation
tonepilot "I'm so frustrated with this broken software!"

Integration Examples

Customer Support Bot:

def handle_customer_message(message):
    engine = TonePilotEngine(mode='gemini', respond=True)
    result = engine.run(message)
    return result['response_text']

Content Writing Assistant:

def get_writing_prompt(topic, desired_tone):
    engine = TonePilotEngine(respond=False)
    result = engine.run(f"Write about {topic} with a {desired_tone} tone")
    return result['final_prompt']

Requirements

• Python 3.8+
• PyTorch (automatically installed)
• Transformers library (automatically installed)
• Internet connection for model downloads on first use

Optional for response generation:

• Google API key (for Gemini mode)

Development

# Clone repository
git clone https://github.com/sdurgi/tonepilot.git
cd tonepilot

# Install in development mode
pip install -e .

# Install development dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Format code
black tonepilot/

Performance

• Emotion Detection: ~50ms on CPU
• Response Generation: 1-3 seconds (depending on model and length)
• Memory Usage: ~500MB (includes cached models)
• Model Downloads: ~1GB on first run (cached locally)

Troubleshooting

Import Errors: Ensure all dependencies are installed with pip install tonepilot

API Key Issues: Verify your .env file or environment variables are set correctly

Model Download Failures: Check internet connection; models download automatically on first use

Memory Issues: Use smaller models or increase available memory

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push to branch (git push origin feature/amazing-feature)
5. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Citation

@software{tonepilot2024,
  title={TonePilot: Emotional Intelligence for Text Generation},
  author={Durgi, Srivani},
  year={2024},
  url={https://github.com/sdurgi/tonepilot}
}

Support

• 📧 Email: [e-mail]
• 🐛 Issues: GitHub Issues

---

Made with ❤️ for building emotionally intelligent AI systems