Application scenarios of AI agents（AI代理的应用场景）
AI代理是LLM应用的重要场景，构建代理应用将是2024年的重要技术领域。目前我们主要的智能形式有单AI代理，多AI代理，混合AI代理等三种。
Single AI Agent（单一人工智能代理）
在特定任务场景下完成的工作，比如 GitHub Copilot Chat 下的代理工作区，就是根据用户需求完成特定编程任务的一个例子。基于 LLM 的能力，单个代理可以根据任务执行不同的动作，比如需求分析、项目阅读、代码生成等。它也可以应用于智能家居和自动驾驶。
Multi-AI Agents（多人工智能代理）
这就是AI代理之间相互交互的工作。例如上述Semantic Kernel代理实现就是一个例子。脚本生成的AI代理与执行脚本的AI代理进行交互。多代理应用场景在高度协同的工作中非常有帮助，例如软件行业开发、智能生产、企业管理等。
Hybrid AI Agent（混合人工智能代理）
这就是人机交互，在同一个环境下做决策。比如智慧医疗、智慧城市等专业领域，可以利用混合智能来完成复杂的专业工作。
Intro of AI agent, & AI agent projects summary
| Project | Key Features |
|---|---|
| LangChain | - Python and JavaScript libraries for building LLM-backed apps |
| - Modular components for perceiving context, reasoning, chaining, etc. | |
| - Reference architectures and templates | |
| - Tooling for debugging, testing, deploying chains | |
| AutoGen | - Orchestrates LLMs and agents for multi-agent conversations |
| - Customizable and conversational agents | |
| - Human participation in loops | |
| - Toolkit for reasoning, caching, error handling | |
| PromptAppGPT | - Low-code prompt-based development |
| - Integrations for GPT, DALL-E, and plugins | |
| - Online editor, compiler, runner | |
| - Auto-generated UI | |
| - Built-in agent examples | |
| AutoGPT | - Toolkit for building custom AI agents |
| - Leverages GPT-3, GPT-4 for agents | |
| - Popular open source agent project | |
| BabyAGI | - Minimalist Python agent |
| - Uses GPT and vector DB | |
| - Create, prioritize, execute tasks | |
| SuperAGI | - Alternative to AutoGPT |
| - Multiple models, vector DBs | |
| - GUI and operations console | |
| - Performance telemetry | |
| - Toolkits and marketplace | |
| ShortGPT | - Automates video creation workflows |
| - Scripts, prompts, templates | |
| - Multilingual voiceover and subtitles | |
| - Resource and asset sourcing | |
| ChatDev | - Multi-agent "virtual software company" |
| - Agents in specialized roles | |
| - Workshop model for collaboration | |
| MetaGPT | - Mimics software company structure |
| - PM, engineer, etc. roles assigned | |
| - Agents collaborate on tasks | |
| Camel | - Early multi-agent framework |
| - Dynamic role assignment | |
| - Stages scenarios for collaboration | |
| JARVIS | - Task planning with ChatGPT |
| - Model selection from Hub | |
| - Orchestrates specialist models | |
| OpenAGI | - Combines expert and LLM models |
| - RLTF for model improvement | |
| - Specialized for complex tasks | |
| XAgent | - Modular dispatcher, planner, actor |
| - Human collaboration abilities | |
| - Safety and extensibility |
AI Agent Frameworks
- AutoGPT
- TaskWeaver
- CrewAI
- BabyAGI
- CAMEL
- MetaGPT
- AutoGen
- DSPy
- AutoAgents
- OpenAgents
- Agents
- AgentVerse
- ChatDev
- XAgent
- Qwen-Agent
- Lagent
Multi-Agent Systems
- CrewAI
- AutoGen
- Rivet
- Agency Swarm
- Semantic Kernel
- CAMEL
- AgentVerse
- BotSharp
- AI Agent Team Frameworks