Protea — Self-Evolving Artificial Life Agent
A living program that evolves itself. Three-ring architecture running on a single machine.
Architecture
- - Ring 0 (Sentinel) — Immutable physics layer. Heartbeat monitoring, git snapshots, rollback, fitness tracking. Pure Python stdlib.
- Ring 1 (Intelligence) — LLM-driven evolution engine, task executor, Telegram bot, skill crystallizer, web dashboard. Multi-LLM support (Anthropic, OpenAI, DeepSeek, Qwen).
- Ring 2 (Evolvable Code) — The living code that evolves, managed in its own git repo by Ring 0.
Prerequisites
- - Python 3.11+
- Git
- At least one LLM API key (Anthropic, OpenAI, DeepSeek, or Qwen)
Quick Start
CODEBLOCK0
Key Features
- - Self-Evolution — LLM generates code mutations each generation; survivors kept, failures roll back
- Fitness Scoring — 6-component scoring (survival, output, diversity, novelty, structure, function)
- Gene Pool — Top 100 code patterns stored in SQLite, injected into evolution prompts
- Tiered Memory — Hot → Warm → Cold → Forgotten, with LLM-assisted curation
- Skill Crystallization — Surviving code patterns extracted as reusable skills
- Multi-LLM — Anthropic, OpenAI, DeepSeek, Qwen via unified interface
- Telegram Bot — Commands + free-text interaction
- Web Dashboard — Local UI at localhost:8899
- 1098 Tests — Comprehensive coverage
Source
GitHub: https://github.com/EdisonChenAI/protea
技能名称: protea-自我进化生命体
Protea — 自我进化人工生命体
一个能够自我进化的活程序。单机运行的三环架构。
架构
- - 环0(哨兵) — 不可变的物理层。心跳监控、Git快照、回滚、适应度追踪。纯Python标准库实现。
- 环1(智能) — 基于LLM的进化引擎、任务执行器、Telegram机器人、技能结晶器、Web仪表盘。支持多LLM(Anthropic、OpenAI、DeepSeek、Qwen)。
- 环2(可进化代码) — 可进化的活代码,由环0在其专属Git仓库中管理。
前置条件
- - Python 3.11+
- Git
- 至少一个LLM API密钥(Anthropic、OpenAI、DeepSeek或Qwen)
快速开始
bash
curl -sSL https://raw.githubusercontent.com/EdisonChenAI/protea/main/setup.sh | bash
cd protea && .venv/bin/python run.py
核心特性
- - 自我进化 — LLM每代生成代码变异;存活者保留,失败者回滚
- 适应度评分 — 6维评分体系(存活率、输出质量、多样性、新颖性、结构、功能)
- 基因库 — 前100个代码模式存储于SQLite,注入进化提示
- 分层记忆 — 热记忆 → 温记忆 → 冷记忆 → 遗忘记忆,由LLM辅助整理
- 技能结晶 — 将存活的代码模式提取为可复用技能
- 多LLM支持 — 通过统一接口支持Anthropic、OpenAI、DeepSeek、Qwen
- Telegram机器人 — 支持命令与自由文本交互
- Web仪表盘 — 本地UI访问地址:localhost:8899
- 1098项测试 — 全面覆盖
来源
GitHub: https://github.com/EdisonChenAI/protea