展示HN：今天早上决定扮演上帝，所以我构建了一个智能体文明。

展示HN：今天早上决定扮演上帝，所以我构建了一个智能体文明。
Show HN: Decided to play god this morning, so I built an agent civilisation

## Werld：一个开放式进化模拟器 Werld是一个完全本地、自包含的计算生态系统，旨在观察开放式进化。它模拟了在Watts-Strogatz小世界网络上进化的自主体，由NEAT风格的大脑和自然选择驱动——没有预定义的目標或人为干预。个体感知、行动、繁殖和死亡，它们的基因组随时间进化。该模拟完全使用Python构建，仅使用标准库，具有可调节的参数，如图形拓扑、季节和29个基因组特征。它支持检查点和安全中断。一个配套的Next.js仪表板（“Werld Observatory”）通过SQLite数据库可视化模拟的进度，跟踪种群动态、大脑复杂性和生态变化。 Werld不需要外部依赖或云服务——只需运行`main.py`并访问`http://localhost:3000`上的仪表板。它专为实验和贡献而设计，并提供完整的技术文档。

一位开发者使用Python构建了“WERLD”，一个开源的人工生命模拟项目，旨在探索具有完全空白神经网络的个体中涌现行为。该项目将30个个体放入一个模拟世界，它们拥有不断进化的神经网络结构、感官输入和可遗传的特征，如沟通和攻击性水平——所有这些都没有预编程的行为或奖励机制。目标是观察从零开始发展出的复杂系统。一个“Werld Observatory”仪表盘提供了种群动态、大脑复杂性以及基于个体之间互动的叙事的实时视图。该创建者希望其他人能够贡献和实验，并指出该项目具有产生有趣故事和见解的潜力。初步反馈集中在改进项目文档（目前由LLM生成）以及在README文件中添加视觉效果上。该项目的GitHub仓库地址是[https://github.com/nocodemf/werld](https://github.com/nocodemf/werld)。

原文

A computational ecosystem where autonomous agents evolve on a graph. No grids, no human-world assumptions — just a Watts-Strogatz small-world network, NEAT-style brains, and natural selection doing the teaching.

Think of it as a digital petri dish: agents perceive, act, reproduce, and die. Their genomes evolve. Brains get more complex (or simpler, if that works better). Communication, memory, and motor patterns are all discoverable — nothing's hardcoded. The goal is open-ended evolution: see what emerges when you remove the training wheels.

Everything runs locally. No cloud, no hosting, no API keys.

Simulation — Pure Python, stdlib only. Graph topology, pheromones, seasons, continuous effectors, evolvable sensory gains, 29 genome traits. Checkpoints, milestones, SIGTERM-safe.
Werld Observatory — Next.js dashboard with 13 sections: population, evolution, brain complexity, ecology, world map, story chapters, etc. Polls SQLite every 4s.

1. Simulation

Python 3.10+. No pip install needed — the sim uses only the standard library.

Runs indefinitely, auto-saves checkpoints to data/. Use --resume to pick up where you left off, --ticks 5000 for a short run, --watchdog to auto-restart on crash.

2. Dashboard

cd dashboard && npm install && npm run dev

Open http://localhost:3000. The dashboard reads from ../data/simulation.db, so start the sim first (or it'll show empty state).

That's it. Sim + dashboard, both local.

├── main.py           # Entry point, CLI, SIGTERM handler
├── config.py         # All tunable params — start here if you want to tweak
├── engine/           # Simulation loop, substrate (graph), story gen
├── agents/           # Genome, cortex, memory, state — the agent stack
├── reasoning/        # NEAT brain (evolvable topology)
├── systems/          # Actions, signals, forking, evolution, entropy
├── persistence/      # SQLite, checkpoints, milestones
├── social/           # X poster (optional, for live instance — not needed to run)
└── dashboard/        # Next.js observatory

CLAUDE.md has the full technical reference — architecture, sensory channels, effector layout, genome traits, everything.

Contributions welcome. See CONTRIBUTING.md for setup, PR process, and what we're looking for.

MIT. See LICENSE.

展示HN：今天早上决定扮演上帝，所以我构建了一个智能体文明。 Show HN: Decided to play god this morning, so I built an agent civilisation

展示HN：今天早上决定扮演上帝，所以我构建了一个智能体文明。
Show HN: Decided to play god this morning, so I built an agent civilisation