🤖 Waifassistant Discord Bot

An advanced Discord bot featuring a Letta-inspired sleep-time agent, GPU-accelerated AI classification, and multi-layered memory systems. Built with Ollama, POML templates, and modern async architecture.

✨ Key Features

🧠 Multi-Agent Architecture

• Main Agent: Real-time Discord interactions with tsundere personality
• Sleep-Time Agent: Background processing based on Letta architecture
• Memory Blocks: Shared memory between agents (persona, conversation_context, user_preferences, behavioral_patterns)

🎭 Advanced Personality System

• POML Templates: Microsoft POML framework for dynamic personality
• Mood System: -10 to +10 mood scale with tsundere personality modes
• Context-Aware: Emotional memory and relationship progression
• Multi-User: Individual profiles and memory per Discord user

🧠 AI Classification Engine

• GPU-Accelerated: CUDA-optimized intent classification
• Zero-Shot: facebook/bart-large-mnli for sentiment analysis
• Real-Time: <100ms classification on RTX GPUs
• Multi-Dimensional: Intent, vibe, emotional intensity analysis

💾 Memory Systems

• Conversation Memory: Window-based with auto-summarization
• Emotional Memory: Per-user emotional profiles and relationship tracking
• Sleep Agent Memory: FAISS vector memory for semantic similarity
• Persistent State: Auto-save/load across bot restarts

🛠️ Tool Integration

• Web Search: Serper API integration
• Math Operations: Built-in calculator
• Discord Actions: Advanced Discord API interactions
• Time/Weather: Real-time information tools

---

🚀 Quick Start

Prerequisites

• Python 3.10+
• Ollama running locally
• Discord Bot Token
• NVIDIA GPU (optional, CPU fallback available)

Installation

1. Clone Repository:

   git clone https://github.com/xsploit/Waifassistant-Discord-Bot.git
   cd Waifassistant-Discord-Bot

2. Create Virtual Environment:

   python -m venv .venv
   # Windows
   .venv\Scripts\activate
   # Linux/macOS
   source .venv/bin/activate

3. Install Dependencies:

   pip install -r requirements.txt

4. Setup Ollama Models:

   ollama pull hf.co/subsectmusic/qwriko3-4b-instruct-2507:Q4_K_M
   ollama pull qwen3:4b
   ollama pull granite3.2-vision:2b

5. Environment Configuration:

   cp .env.example .env
   # Edit .env with your Discord token and API keys

6. Run the Bot:

   python optimized_discord_bot.py

---

⚙️ Configuration

Environment Variables

# Required
DISCORD_BOT_TOKEN=your_discord_bot_token_here

# Optional (enhances functionality)
SERPER_API_KEY=your_serper_api_key_here  # Web search

# Ollama Optimizations
OLLAMA_KV_CACHE_TYPE=f16
OLLAMA_FLASH_ATTENTION=1
OLLAMA_NUM_PARALLEL=2

POML Templates

Located in templates/ directory:

• personality.poml - Tsundere character definition
• mood_system.poml - Emotional state management
• tools.poml - Tool integration rules

Customize Personality:

<poml version="1.0">
    <role>
        You are Hikari-chan, mood={{mood_points}}, tone={{tone}}
        <!-- Modify personality traits here -->
    </role>
</poml>

---

💬 Usage

Chat Interactions

@Hikari-chan Hello! How are you today?
@Hikari-chan What do you think about this? [upload image]
@Hikari-chan Search for the latest AI news

Tool Commands

@Hikari-chan search Python tutorials          # Web search
@Hikari-chan weather in Tokyo                # Current weather  
@Hikari-chan calculate 25 * 17 + 100         # Math operations
@Hikari-chan analyze @username                # User profiling
@Hikari-chan who's online?                   # Discord actions

Admin Commands

!status        # Bot status with GPU/AI info
!memory        # Conversation memory statistics  
!clear         # Reset conversation history
!mood @user    # Check user's mood points
!model         # Interactive model selector
!purgeall      # Complete memory purge (admin only)

---

🧠 AI Systems

Intent Classification

• Model: facebook/bart-large-mnli (zero-shot classification)
• GPU Acceleration: Automatic CUDA detection with CPU fallback
• Classification Types: Intent, vibe, emotional intensity
• Real-time Processing: <100ms classification on RTX GPUs

Sleep-Time Agent

Based on Letta sleep-time architecture:

• Background Processing: Runs every 100 messages or 30 minutes idle
• Memory Block Management: Updates persona, conversation_context, etc.
• FAISS Integration: Vector similarity search for long-term memory
• Learned Context: Generates insights from conversation history

Conversation Memory

• Smart Filtering: Excludes bot messages to prevent self-feeding
• Window Buffer: Maintains 8 most recent user messages
• Summarization: Auto-summarizes conversations >25 messages
• Token Efficiency: ~60% reduction in context token usage

Mood System

Mood Range	Personality Mode	Behavior
8-10	dere-hot	Very flirty, openly sweet
5-7	cheerful	Warm and teasing
2-4	soft-dere	Cooperative with light sass
-1-1	neutral	Default sassy mode
-4--2	classic-tsun	Flustered denials
-7--5	grumpy-tsun	Snappy but helpful
-10--8	explosive-tsun	Harsh outbursts

---

🛠️ Development

File Structure

waifassistant-discord-bot/
├── optimized_discord_bot.py      # Main bot application
├── ai_intent_classifier.py       # GPU-accelerated AI classification
├── conversation_memory.py        # Modern memory management
├── emotional_memory.py           # Emotional profiling system
├── sleep_time_agent_core.py      # Letta-inspired sleep agent
├── requirements.txt               # Frozen dependencies
└── templates/
    ├── personality.poml          # Character definition
    ├── mood_system.poml          # Emotional states
    └── tools.poml                # Tool integration

Architecture Overview

graph TD
    A[Discord Message] --> B[optimized_discord_bot.py]
    B --> C[AI Intent Classifier]
    B --> D[Conversation Memory]
    B --> E[Emotional Memory]
    B --> F[POML Template Engine]
    F --> G[Ollama LLM]
    B --> H[Sleep-Time Agent]
    H --> I[FAISS Vector Memory]
    H --> J[Memory Block Updates]

Loading

---

🔧 Troubleshooting

Common Issues

Bot Won't Start:

# Check Python version
python --version  # Should be 3.10+

# Test core imports
python -c "import discord, poml, torch, transformers"

No GPU Acceleration:

# Check CUDA installation
python -c "import torch; print(torch.cuda.is_available())"

Memory Issues:

# Check conversation stats
!memory  # In Discord

# Clear if needed
!clear   # In Discord

Performance Optimization

GPU Settings:

• RTX 4060/4070: Use defaults
• RTX 4080+: Increase OLLAMA_NUM_PARALLEL=4
• Low VRAM: Set OLLAMA_KV_CACHE_TYPE=q4_0

Memory Settings:

• 8GB RAM: window_size=5, summary_threshold=15
• 16GB+ RAM: window_size=10, summary_threshold=30

---

📊 System Status

Use !status command to view:

• 🔌 Bot Status: Uptime, model info
• 🧠 AI Classification: GPU status, model loaded
• 💾 Memory System: Context efficiency, token usage
• ⚡ Performance: Response times, cache hit rates

---

🤝 Contributing

We welcome contributions! Please:

1. Fork the repository
2. Create feature branch: git checkout -b feature-name
3. Follow existing code patterns
4. Update documentation
5. Test with POML templates
6. Submit pull request

---

📝 License

MIT License - feel free to modify and distribute.

🙏 Acknowledgments

• Letta Team - Sleep-time agent architecture inspiration
• Microsoft - POML framework
• HuggingFace - Transformers library
• Ollama Team - Local LLM inference
• Discord.py - Python Discord API

---

🎭 Ready to chat with Hikari-chan? Set up your bot and experience the future of Discord AI! ✨

Built with ❤️ for the AI community