DieLoli is an ASCII art-based ERALike game featuring thousands of freely simulated NPCs, offering an open-ended sandbox experience. It targets players seeking emergent gameplay and allows integration with large language models (LLMs) for enhanced NPC interactions, though this feature is experimental and resource-intensive.

How It Works

The game simulates thousands of NPCs with free computation, akin to RimWorld or Dwarf Fortress, leading to unique playthroughs. Its "AI algorithm" differs from typical AIGC, with technical discussions available via a linked post. LLM integration supports Ollama or external API modes, allowing players to use models like DeepSeek-R1-1.5B, with configurable prompts and API endpoints.

Quick Start & Requirements

• Windows/macOS: Download the latest release, extract, and run game.exe.
• Linux: Download source, install dependencies from requirements.txt via pip, and run game.py.
• Prerequisites: LLM integration requires Ollama installation and a compatible GPU (e.g., RTX 3060 for 20s inference with DeepSeek-R1-1.5B).
• Resources: Initial run may require significant data preheating. LLM features can cause game stuttering.
• Links: Download: 本地址, Discord: https://discord.gg/Hu67GuXkfV

Highlighted Details

• Features thousands of freely simulated NPCs for emergent gameplay.
• Supports LLM integration via Ollama or external APIs for enhanced NPC interactions.
• Developed using the pyera engine, which also powers the erArk game.
• Includes experimental features in a separate "Test Version".

Maintenance & Community

• Active development with stable and test versions.
• Community discussions and development plans are shared on Discord.
• Contact: pokemonchw@gmail.com.

Licensing & Compatibility

• License: CC BY-NC-SA.
• Restrictions: Non-commercial use only. Requires attribution and same-clause distribution if modified.

Limitations & Caveats

The game contains mature themes (sexual, violent, religious content) for gameplay and artistic reasons. LLM integration is experimental, may cause performance issues (stuttering, long inference times), and is not recommended for general use yet.