AI Agent
一个使用工具进行规划并采取行动，而不仅仅是回答的系统。
概览
AI Agent 将语言模型包裹在一个循环中：它决定下一步做什么、调用工具、观察结果，并持续这一过程直到任务完成。它解决了单个模型响应无法查询信息、运行代码或改变状态的局限。它之所以存在，是因为许多有用的任务是多步骤的，并且依赖于模型所不具备的信息，而真正的难点不在于模型，而在于控制循环、限制成本以及从失败的步骤中恢复。
技术图谱
AI Agent 如何与技术栈的其他部分相连。
使用与依赖
- LLM concept Models AI Agent uses LLM High ✓ source
- LangChain framework LLM Frameworks AI Agent uses LangChain Medium ✓ source
- RAG technique Techniques AI Agent uses RAG Medium ✓ source
应用于
- Agent Memory technique AI Agents Agent Memory used for AI Agent High ✓ source
- AutoGen framework Frameworks AutoGen used for AI Agent High ✓ source
- Constrained Decoding technique Techniques Constrained Decoding used for AI Agent High ✓ source
- Context Engineering technique Prompting Context Engineering used for AI Agent High ✓ source
- CrewAI framework Frameworks CrewAI used for AI Agent High ✓ source
- DSPy framework Frameworks DSPy used for AI Agent High ✓ source
- Function Calling technique Prompting Function Calling used for AI Agent High ✓ source
- Guardrails technique Safety & Alignment Guardrails used for AI Agent Medium ✓ source
- LangSmith tool Evaluation LangSmith used for AI Agent High ✓ source
- Prompt Caching technique Infrastructure Prompt Caching used for AI Agent High ✓ source
- ReAct technique AI Agents ReAct used for AI Agent High ✓ source
- SWE-bench benchmark Benchmarks SWE-bench used for AI Agent High ✓ source
- Structured Output technique Prompting Structured Output used for AI Agent High ✓ source