MiniAGI
MiniAGI is a simple general-purpose AI agent based on the OpenAI API.
No commits in 40 months — may not be actively maintained. See maintained alternatives →
open-sourceagent-frameworks
2.9k
Stars
+2
Stars/month
0
Commits (90d)
0
Releases (6m)
Star Growth
+14 (0.5%)
Overview
MiniAGI是一个基于OpenAI API的简单通用AI代理,旨在实现自主任务执行。它结合了强大的提示工程、最小化工具集、思维链推理和带摘要的短期记忆功能。该代理具备内在独白和自我批评能力,可以通过Python和Shell命令执行来完成复杂任务。MiniAGI兼容GPT-3.5-Turbo和GPT-4模型,提供了一个轻量级但功能完整的AGI实现框架。其设计哲学是用最少的组件实现最大的自主性,适合研究者和开发者探索AI代理的核心机制。工具包含可选的批评模块来提高准确性,以及完整的任务执行和反思循环。
Deep Analysis
Key Differentiator
Minimal autonomous agent with self-criticism and inner monologue, achieving complex tasks with a deliberately small codebase
⚡ Capabilities
- • autonomous-agent
- • chain-of-thought
- • self-criticism
- • inner-monologue
- • python-execution
- • web-browsing
🔗 Integrations
openai-gpt-3.5-turboopenai-gpt-4
✓ Best For
- ✓ autonomous-task-experimentation
- ✓ learning-agent-architecture
- ✓ simple-automation-tasks
✗ Not Ideal For
- ✗ production-systems
- ✗ safety-critical-applications
- ✗ complex-multi-step-workflows
Languages
python
Deployment
clilocal
⚠ Known Limitations
- ⚠ safety-risks-from-code-execution
- ⚠ high-api-costs-with-critic
- ⚠ no-gui
- ⚠ limited-tool-set
Pros
- + 轻量级设计,代码简洁易于理解和修改,适合学习AGI架构原理
- + 内置工具执行能力,可以运行Python代码和Shell命令来完成实际任务
- + 具备自我批评和反思机制,可选的critic模块提升任务执行准确性
Cons
- - 功能相对基础,缺乏复杂的规划和多步骤任务管理能力
- - 依赖OpenAI API,在网络连接或API配额方面存在限制
- - 安全风险警告明确,可能生成有害命令,需要谨慎使用
Use Cases
- • 自动化编程任务,如生成图像、创建网站或数据处理脚本
- • 研究和教育用途,学习AI代理的基本架构和实现原理
- • 原型开发,快速测试基于LLM的自主任务执行想法
Getting Started
1. 克隆仓库并安装依赖:git clone repo, pip install -r requirements.txt;2. 配置环境:复制.env_example为.env并添加OpenAI API密钥;3. 运行任务:python miniagi.py "你的目标描述"
Alternatives
B
BabyAGI
A
AutoGPT
AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.
l
loopgpt
Modular Auto-GPT Framework
X
XAgent
An Autonomous LLM Agent for Complex Task Solving
Compare MiniAGI
Maintain MiniAGI?
Show your live rank in your README, or put MiniAGI in front of every visitor to AgentoolRank.