达尔文-哥德尔机器
经 AI Skill Hub 精选评估，达尔文-哥德尔机器 获评「推荐使用」。这款Agent工作流在功能完整性、社区活跃度和易用性方面表现出色，AI 评分 7.5 分，适合有一定技术背景的用户使用。
📚 深度解析
达尔文-哥德尔机器 工作流的设计遵循"最小配置，最大复用"原则：核心逻辑已经封装好，用户只需配置自己的 API Key 和业务参数即可快速上手。工作流内置错误处理和重试机制，在网络波动或 API 限速等情况下仍能稳定运行，适合作为生产环境的自动化基础设施。
在实际部署时，建议先在测试环境中运行 3-5 次，验证各个环节的输出结果符合预期，再部署到生产环境。AI Skill Hub 评分 7.5 分，是同类 Agent 工作流中的精选推荐。
📋 工具概览
自我改进AI代理，实现达尔文-哥德尔机器研究论文
达尔文-哥德尔机器 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排，将复杂的多步骤任务拆解为清晰的自动化流程，实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成，适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
📖 中文文档
自我改进AI代理，实现达尔文-哥德尔机器研究论文
达尔文-哥德尔机器 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排，将复杂的多步骤任务拆解为清晰的自动化流程，实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成，适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
- 可视化 Agent 工作流编排，无需编写复杂代码
- 支持多步骤自动化任务链，实现全流程无人值守
- 与外部 API、数据库和第三方服务无缝集成
- 内置错误处理与自动重试机制，保障稳定运行
- 提供可复用的自动化模板，快速在同类场景部署
- 自动化日常重复性工作，将精力集中于创造性任务
- 构建数据采集 → 处理 → 输出的完整自动化管线
- 实现跨平台、跨系统的数据流转和业务协同
# 方式一：pip 安装（推荐）
pip install darwin-godel-machine
# 方式二：虚拟环境安装（推荐生产环境）
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install darwin-godel-machine
# 方式三：从源码安装（获取最新功能）
git clone https://github.com/lemoz/darwin-godel-machine
cd darwin-godel-machine
pip install -e .
# 验证安装
python -c "import darwin_godel_machine; print('安装成功')"
- 访问 GitHub 仓库获取工作流文件
- 在对应平台（Dify / Flowise / Make 等）中找到「导入工作流」功能
- 上传工作流文件
- 按照提示配置必要的环境变量和 API Key
- 运行测试确认流程正常后投入使用
# 命令行使用
darwin-godel-machine --help
# 基本用法
darwin-godel-machine input_file -o output_file
# Python 代码中调用
import darwin_godel_machine
# 示例
result = darwin_godel_machine.process("input")
print(result)
# darwin-godel-machine 配置文件示例（config.yml） app: name: "darwin-godel-machine" debug: false log_level: "INFO" # 运行时指定配置文件 darwin-godel-machine --config config.yml # 或通过环境变量配置 export DARWIN_GODEL_MACHINE_API_KEY="your-key" export DARWIN_GODEL_MACHINE_OUTPUT_DIR="./output"
Darwin Gödel Machine
A Self-Improving AI System for Evolutionary Code Enhancement
🧬 Overview
The Darwin Gödel Machine (DGM) is an innovative implementation of self-improving AI agents that iteratively modify their own Python codebase to enhance their coding capabilities. Unlike traditional approaches that rely on formal proofs, DGM uses empirical validation through coding benchmarks to drive evolutionary improvement.
Based on the research paper "Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents" (arXiv:2505.22954), this implementation demonstrates how AI systems can achieve self-referential self-improvement through population-based exploration and empirical validation.
For more details, see the official blog post from Sakana AI.
🔑 Key Features
- 🔄 Self-Referential Self-Improvement: Agents modify their own source code to improve performance
- 📊 Empirical Validation: Changes validated through coding benchmarks rather than formal proofs
- 🏗️ Population-Based Evolution: Maintains archive of all valid agents for diverse exploration
- 🤖 Foundation Model Integration: Supports Claude, Gemini, and OpenAI models
- 🛠️ Tool-Equipped Agents: Agents use Bash and file editing tools to solve problems
- 🏪 Agent Archive System: Stores successful agents with performance and novelty metrics
- 🎯 Benchmark-Driven Evolution: Uses custom coding challenges to measure improvement
- 🔒 Guarded Execution: Workspace-scoped file access, command filtering, and hard timeouts with process-group cleanup (Docker isolation is planned, not yet implemented)
Agent Capabilities
Each agent is a complete coding system that can: - Solve Coding Problems: Use LLM reasoning to understand and solve tasks - Use Tools: Execute bash commands and edit files - Self-Analyze: Review its own performance and identify weaknesses - Self-Modify: Propose and implement improvements to its own code - Maintain Validity: Preserve its core capabilities while evolving
🔒 Safety Features
- Workspace Containment: File edits and bash redirects are resolved and confined to the agent's workspace; path-traversal escapes are rejected
- Command Filtering: Dangerous commands are blocked before execution
- Hard Timeouts: Benchmark and tool subprocesses are killed (entire process group) on timeout
- Validation Checks: Modified agents must parse, define a working Agent class, and load before admission to the archive
- Human Oversight: Optional pause points for review
- Comprehensive Logging: Full audit trail of all changes
Note: execution is guarded but not fully isolated — Docker-based sandboxing is planned (sandbox/ contains the stub). Run untrusted evolution experiments inside a container or VM.
Prerequisites
- Python 3.9+
- API keys for supported Foundation Models (Claude, Gemini, or OpenAI)
Install dependencies (includes pytest)
pip install -r requirements.txt
Installation
1. Clone the repository:
git clone https://github.com/lemoz/darwin-godel-machine.git
cd darwin-godel-machine2. Install dependencies:
pip install -r requirements.txt3. Configure API keys: ```bash cp .env.example .env
Development Setup
```bash
🚀 Quick Start
Benchmark Examples
The system includes several coding challenges:
```yaml
Edit .env with your API keys
4. **Run the system:**🔧 Configuration
The system uses YAML configuration files to control behavior:
```yaml
config/dgm_config.yaml
fm_providers: primary: anthropic # or 'gemini', 'openai' anthropic: model: claude-sonnet-4-6 api_key: ${ANTHROPIC_API_KEY} gemini: model: gemini-2.5-flash-preview-05-20 api_key: ${GEMINI_API_KEY}
dgm_settings: max_iterations: 100 pause_after_iteration: true sandbox_timeout: 300
benchmarks: enabled: - string_manipulation - list_processing - simple_algorithm ```
Run with default settings (3 generations)
python run_dgm.py
Verify message formatting and provider config (no network needed)
python -m pytest tests/unit/test_fm_connection.py
Dump full agent conversations to debug/ (set in config/dgm_config.yaml
Run the full test suite (155 tests, no API keys needed)
python -m pytest
📚 References
Core Components
#### 1. DGM Controller (dgm_controller.py) Orchestrates the main evolution loop: - Parent selection from archive - Self-modification coordination - Benchmark evaluation - Archive management
#### 2. Agent System (agent/) LLM-powered coding agents with: - Foundation Model integration - Tool usage capabilities (Bash, File editing) - Task solving and self-modification abilities
#### 3. Archive Management (archive/) - Agent Archive: Stores every valid agent (unbounded, per the paper) with full lineage - Parent Selector: Implements the paper's selection rule — sigmoid-scaled performance times a 1/(1+children) exploration bonus, sampled categorically - Lineage Visualization: Generate an SVG or HTML family tree from archive metadata
python scripts/generate_archive_lineage.py --archive-dir archive/agents --output docs/archive-lineage.html
#### 4. Evaluation System (evaluation/) - Benchmark Runner: Executes agents on coding challenges - Validator: Ensures agents compile and maintain capabilities - Scorer: Calculates performance metrics
#### 5. Self-Modification (self_modification/) - Diagnosis: Analyzes agent performance issues - Proposal: Generates improvement suggestions - Implementation: Applies code modifications
Testing Components
```bash
Test specific components
python -m pytest tests/unit/test_agent.py python -m pytest tests/integration/ # includes a full no-network DGM generation ```
🐛 Troubleshooting
高质量的AI工作流自动化项目
⚡ 核心功能
- 可视化 Agent 工作流编排，无需编写复杂代码
- 支持多步骤自动化任务链，实现全流程无人值守
- 与外部 API、数据库和第三方服务无缝集成
- 内置错误处理与自动重试机制，保障稳定运行
- 提供可复用的自动化模板，快速在同类场景部署
👥 适合人群
🎯 使用场景
- 自动化日常重复性工作，将精力集中于创造性任务
- 构建数据采集 → 处理 → 输出的完整自动化管线
- 实现跨平台、跨系统的数据流转和业务协同
⚖️ 优点与不足
- +大幅减少重复性人工操作
- +可视化流程，清晰直观
- +可扩展性强，支持复杂场景
- −未明确开源协议，商用场景需谨慎评估
- −初始配置和调试需投入一定时间
- −强依赖外部服务的稳定性
- −复杂场景需具备一定技术基础
该工具未明确声明开源协议，商业使用前请联系原作者确认授权范围，避免侵权风险。
AI Skill Hub 为第三方内容聚合平台，本页面信息基于公开数据整理，不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后，再部署至生产环境，并做好必要的安全评估。
🔗 相关工具推荐
❓ 常见问题 FAQ
AI Skill Hub 点评：达尔文-哥德尔机器 的核心功能完整，质量良好。对于自动化工程师和运维人员来说，这是一个值得纳入个人工具库的选择。建议先在非生产环境试用，再逐步推广。
| 原始名称 | darwin-godel-machine |
| Topics | AI自我改进达尔文-哥德尔机器 |
| GitHub | https://github.com/lemoz/darwin-godel-machine |
| 语言 | Python |
收录时间：2026-06-12 · 更新时间：2026-06-12 · License：未公布 · AI Skill Hub 不对第三方内容的准确性作法律背书。