A Discord bot that stores long-term chat history using SQLite database with RAG (Retrieval-Augmented Generation) capabilities powered by custom embedding models.

Vibe Discord Bot with RAG Chat History

Quick Start - Available Commands

Pre-built Bots

Command	Description	Example Usage
`!doodlebob`	Generate images from text	`!doodlebob a cat sitting on a moon`
`!retcon`	Edit images with text prompts	`!retcon <image attachment> Make it sunny`

Custom Bot Management

Command	Description	Example Usage
`!custom <name> <personality>`	Create a custom bot with specific personality	`!custom alfred you are a proper british butler`
`!list-custom-bots`	List all available custom bots	`!list-custom-bots`
`!delete-custom-bot <name>`	Delete your custom bot	`!delete-custom-bot alfred`

Using Custom Bots

Once you create a custom bot, you can interact with it directly by prefixing your message with the bot name:

!<bot_name> <your message>

Example:

Create a bot: !custom alfred you are a proper british butler
Use the bot: alfred Could you fetch me some tea?
The bot will respond in character as a British butler

Features

Long-term chat history storage: Persistent storage of all bot interactions
RAG-based context retrieval: Smart retrieval of relevant conversation history using vector embeddings
Custom embedding model: Uses qwen3-embed-4b for semantic search capabilities
Efficient message management: Automatic cleanup of old messages based on configurable limits
Long-term chat history storage: Persistent storage of all bot interactions
RAG-based context retrieval: Smart retrieval of relevant conversation history using vector embeddings
Custom embedding model: Uses qwen3-embed-4b for semantic search capabilities
Efficient message management: Automatic cleanup of old messages based on configurable limits

Setup

Prerequisites

Python 3.10 or higher
uv package manager
Embedding API key
Discord bot token

Environment Variables

Create a .env file or export the following variables:

# Discord Bot Token
export DISCORD_TOKEN=your_discord_bot_token

# Embedding API Configuration
export OPENAI_API_KEY=your_embedding_api_key
export OPENAI_API_ENDPOINT=https://llama-embed.reeselink.com/embedding

# Image Generation (optional)
export IMAGE_GEN_ENDPOINT=http://toybox.reeselink.com:1234/v1
export IMAGE_EDIT_ENDPOINT=http://toybox.reeselink.com:1235/v1

# Database Configuration (optional)
export CHAT_DB_PATH=chat_history.db
export EMBEDDING_MODEL=qwen3-embed-4b
export EMBEDDING_DIMENSION=2048
export MAX_HISTORY_MESSAGES=1000
export SIMILARITY_THRESHOLD=0.7
export TOP_K_RESULTS=5

Installation

Sync dependencies with uv:

uv sync

Run the bot:

uv run main.py

How It Works

Database Structure

The system uses two SQLite tables:

chat_messages: Stores message metadata
- message_id, user_id, username, content, timestamp, channel_id, guild_id
message_embeddings: Stores vector embeddings for RAG
- message_id, embedding (as binary blob)

RAG Process

When a message is received, it's stored in the database
An embedding is generated using OpenAI's embedding API
The embedding is stored alongside the message
When a new message is sent to the bot:
- The system searches for similar messages using vector similarity
- Relevant context is retrieved and added to the prompt
- The LLM generates a response with awareness of past conversations

Configuration Options

MAX_HISTORY_MESSAGES: Maximum number of messages to keep (default: 1000)
SIMILARITY_THRESHOLD: Minimum similarity score for context retrieval (default: 0.7)
TOP_K_RESULTS: Number of similar messages to retrieve (default: 5)
EMBEDDING_MODEL: OpenAI embedding model to use (default: text-embedding-3-small)

Usage

The bot maintains conversation context automatically. When you ask a question, it will:

Search for similar past conversations
Include relevant context in the prompt
Generate responses that are aware of the conversation history

File Structure

vibe_discord_bots/
├── main.py              # Main bot application
├── database.py          # SQLite database with RAG support
├── pyproject.toml       # Project dependencies (uv)
├── .env                 # Environment variables
├── .venv/               # Virtual environment (created by uv)
└── README.md           # This file

Build

Using uv

# Set environment variables
export DISCORD_TOKEN=$(cat .token)
export OPENAI_API_KEY=your_api_key
export OPENAI_API_ENDPOINT="https://llama-cpp.reeselink.com"
export IMAGE_GEN_ENDPOINT="http://toybox.reeselink.com:1234/v1"
export IMAGE_EDIT_ENDPOINT="http://toybox.reeselink.com:1235/v1"

# Run with uv
uv run main.py

Container

# Build
podman build -t vibe-bot:latest .

# Run
podman run --env-file .env localhost/vibe-bot:latest

Docs

Open AI

Chat

https://developers.openai.com/api/reference/resources/chat/subresources/completions/methods/create

Images

https://developers.openai.com/api/reference/python/resources/images/methods/edit

Models

Qwen3.5

We recommend using the following set of sampling parameters for generation

Non-thinking mode for text tasks: temperature=1.0, top_p=1.00, top_k=20, min_p=0.0, presence_penalty=2.0, repetition_penalty=1.0
Non-thinking mode for VL tasks: temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
Thinking mode for text tasks: temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
Thinking mode for VL or precise coding (e.g. WebDev) tasks : temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0

Please note that the support for sampling parameters varies according to inference frameworks.