PuddingBot with Ollama Integration

PuddingBot is a Discord bot designed for the Google Snake gaming community, now featuring local AI processing using Ollama instead of external APIs.

Features

• Discord Integration: Full Discord bot functionality with message handling
• Local AI Processing: Uses Ollama for AI responses (no external API dependencies)
• Google Snake Expertise: Extensive knowledge about Google Snake mechanics, speedrunning, and community
• Wall Pattern Solver: Advanced algorithm to solve wall patterns in Google Snake
• GIF Responses: KLIPY API integration for animated responses
• Channel Management: Special handling for specific Discord channels

Prerequisites

• Docker installed on your system
• Discord bot token (set in environment variables)

Quick Start

1. Set up environment variables: Create a .env file in the project root:

   DISCORD_TOKEN=your_discord_bot_token_here
   KLIPY_KEY=your_klipy_api_key_here
   

2. Build and run with Docker:

   # From repo root (or use the helper scripts)
   ./scripts/run_docker.sh
   # Windows: scripts\run_docker.bat
   
   # Or manually:
   docker build -t puddingbot .
   docker run -d --name puddingbot-container --env-file .env puddingbot

3. Check logs:

   docker logs puddingbot-container

Docker Setup

The Docker container includes:

• Ubuntu 22.04 base image
• Ollama for local AI processing
• Python 3 with all required dependencies
• qwen3:0.6b model (automatically downloaded on start; override with OLLAMA_MODEL)

AI Integration

The bot now uses Ollama for AI responses instead of external APIs:

• Model: qwen3:0.6b (fastest current tools-capable Qwen3; thinking disabled for latency)
• Local Processing: All AI responses are generated locally
• No External Dependencies: No need for OpenAI API keys or external services

Commands

• @PuddingBot <message> - Get AI response
• @PuddingBot clear context - Clear conversation context
• gif <emotion> - Get a random GIF
• roll dice - Roll a 6-sided die
• pattern <pattern_string> - Solve a wall pattern (pudding clipboard paste works as-is)
• /wallall - Same Wall All solver via slash command

File Structure

/
  main.py, start.sh, Dockerfile, requirements.txt
  data_management.py, github_cache_fetcher.py
  chat/          Message replies and Ollama AI
  cogs/          Discord slash/context command extensions
  wall/          Wall All solver, renderer, Discord stream updates
  native/        C Warnsdorff DFS extension (warnsdorff_c)
  tests/         Local smoke tests
  scripts/       Docker helper scripts
  assets/        Memes, fonts, GIFs

Wall All path search uses the native C DFS when warnsdorff_c is installed (built in Docker; falls back to pure Python otherwise). Force Python with HAMPATH_DFS=python.

Troubleshooting

1. Ollama not starting: Check Docker logs for Ollama startup issues
2. Model not found: The build process should automatically download the model
3. Discord connection issues: Verify your Discord token is correct
4. Memory issues: The qwen3:0.6b model needs roughly ~1GB RAM

Development

To test the Ollama integration locally:

python3 tests/test_ollama.py

FastSnakeStats cache smoke test:

python3 tests/test_fastsnakestats.py

Notes

• The bot will automatically start Ollama and wait for it to be ready before starting the Discord bot
• All AI responses are generated locally, so no internet connection is required for AI functionality
• The container includes proper error handling and graceful shutdown