Discord AI Chatbot

A sophisticated Discord bot that leverages large language models (LLMs) for creating immersive, context-aware conversations with long-term memory capabilities.

🌟 Features

• Long-term memory - Remembers past conversations using semantic search
• Context awareness - Maintains conversational context across messages
• Freewill mode - Bot can initiate conversations after periods of inactivity
• Multiple LLM support - Compatible with Gemini, OpenAI, Claude, and other providers
• Docker ready - Easy deployment with Docker and docker-compose
• Custom roleplay guidelines - Configurable personality and interaction styles
• Temporal awareness - Acknowledges time gaps between messages

📋 Prerequisites

• Rust (latest stable recommended)
• Discord bot token
• API keys for your preferred LLM provider (Gemini, OpenAI, etc.)
• Docker and docker-compose (optional, for containerized deployment)

🚀 Quick Start

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/chatbot.git
   cd chatbot

2. Copy the example config and customize it:

   cp config.example.toml config.toml
   # Edit config.toml with your API keys and preferences

3. Build the project:

   cargo build --release

Supported LLM Providers

This project depends on my dependency rig-dyn for dynamic LLM provider integration with the rig crate. Currently, the following LLM providers are supported:

• anthropic
• azure
• cohere
• deepseek
• galadriel
• gemini
• groq
• huggingface
• hyperbolic
• mira
• moonshot
• openai
• openrouter
• ollama
• perplexity
• together
• xai

Running

Using Cargo

cargo run --release

Using Docker

docker-compose up -d

⚙️ Configuration

The bot is configured via the config.toml file. Key configuration sections include:

Discord Config

[config.discord]
token = "YOUR_DISCORD_BOT_TOKEN"

LLM Config

[config.llm]
use_tools = true
force_lowercase = true
similarity_threshold = 0.5

[config.llm.completion]
model = "gemini-2.0-flash-thinking-exp-01-21"
provider = "gemini"
api_key = "YOUR_API_KEY"

Memory Configuration

Configure memory storage parameters:

[config.llm.embedding]
model = "text-embedding-004"
provider = "gemini"
api_key = "YOUR_API_KEY"
qdrant_host = "127.0.0.1"
qdrant_port = 6334
qdrant_https = false

Bot Personality

[config.context.system]
chatbot_name = "your_bot_name"
user_name = "default_user_name"
about = "Description of your bot's personality and background"
max_ltm = 100
age = "19 years old"

📚 Architecture

The bot consists of several key components:

• Discord Integration: Manages Discord events and message handling
• Chat Engine: Core conversational logic
• Memory System: Long-term and short-term memory management
  • Uses Qdrant vector database for semantic search
  • Stores conversation summaries for future recall
• LLM Client: Interfaces with various LLM providers

🔧 Commands

• /clear - Clear conversation history
• /config - Update configuration settings
• /reload - Reload the bot configuration

🤖 Memory Management

The bot uses a two-tiered memory system:

1. Short-term memory (STM): Recent conversation history kept in active memory
2. Long-term memory (LTM): Important information extracted and stored in a vector database for semantic search and recall

When the bot needs to recall information, it:

1. Embeds the current query using the embedding model
2. Searches the vector database for semantically similar memories
3. Incorporates relevant memories into its response context

🔄 Freewill Mode

The bot can initiate conversations after periods of inactivity:

• Configurable minimum and maximum wait times
• Probability increases exponentially with time since last interaction
• Summarizes conversation context before initiating new interactions

📝 License

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

🙏 Acknowledgements

• Built with Rust, Serenity, and Qdrant
• Uses rig for LLM integration