Agent论文和项目大合集

综述

Agent AI: Surveying the Horizons of Multimodal Interaction斯坦福大学、微软研究院、加州大学洛杉矶分校、华盛顿大学等

• https://arxiv.org/pdf/2401.03568

The Rise and Potential of Large Language Model Based Agents: A Survey复旦大学自然语言处理团队、米哈游公司

• https://xuanjing-huang.github.io/files/agent.pdf
• https://github.com/woooodyy/llm-agent-paper-list

A survey on large language model based autonomous agents中国人民大学高瓴人工智能学院研究团队

• https://link.springer.com/content/pdf/10.1007/s11704-024-40231-1.pdf
• https://github.com/paitesanshi/llm-agent-survey
• https://github.com/melih-unsal/demogpt

Large Language Model based Multi-Agents: A Survey of Progress and Challenges南方科技大学等

• https://arxiv.org/pdf/2402.01680
• https://github.com/taichengguo/llm_multiagents_survey_papers

Personal LLM Agents: Insights and Survey about the Capability, Efficiency and Security清华大学人工智能产业研究院

• https://arxiv.org/pdf/2401.05459
• https://github.com/mobilellm/personal_llm_agents_survey

Understanding the planning of LLM agents: A survey中国科学技术大学、华为诺亚方舟实验室

• https://arxiv.org/pdf/2402.02716

Toward Efficient Agents: A Survey of Memory, Tool learning, and Planning高效Agents技术全面综述 上海AI Lab & 复旦&中科院&上交大等9所高校

• https://arxiv.org/abs/2601.14192
• https://efficient-agents.github.io/
• https://github.com/yxf203/Awesome-Efficient-Agents

Memory in the LLM Era: Modular Architectures and Strategies in a Unified FrameworkLLM Agent统一记忆框架综述

• https://arxiv.org/abs/2604.01707
• https://github.com/Yanchen398/Memory-in-the-LLM-Era

Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness EngineeringAgents统一综述：Harness、记忆、Skills和协议 上交大、中山大学、卡梅隆等发表长

• https://arxiv.org/pdf/2604.08224

Graph-based Agent Memory: Taxonomy, Techniques, and ApplicationsGraph增强Agent记忆综述

• https://arxiv.org/pdf/2602.05665
• https://github.com/DEEP-PolyU/Awesome-GraphMemory

Rethinking Memory Mechanisms of Foundation Agents in the Second Half: A Survey基础智能体记忆机制综述

• https://arxiv.org/pdf/2602.06052

The Path Ahead for Agentic AI：Challenges and OpportunitiesAgentic AI综述

• https://huggingface.co/papers/2601.02749

AI Agent Systems: Architectures, Applications, and EvaluationAgentic AI综述

• http://arxiv.org/abs/2601.01743v1

Agent 核心系统形态（按智能体数量划分）

这类是 Agent 最基础的分类维度，是入门时首先要理解的核心概念，直接决定系统的协作模式与应用场景。

单智能体：独立完成任务的单一智能体系统，聚焦 "个体决策、自主执行"，如个人助理、单机器人控制。

单智能体 21

Font-Agent: Enhancing Font Understanding with Large Language Models

• https://openaccess.thecvf.com/content/CVPR2025/papers/Lai_Font-Agent_Enhancing_Font_Understanding_with_Large_Language_Models_CVPR_2025_paper.pdf

ATA: Adaptive Transformation Agent for Text-Guided Subject-Position Variable Background Inpainting

• https://arxiv.org/abs/2504.01603

Feature4X: Bridging Any Monocular Video to 4D Agentic AI with Versatile Gaussian Feature Fields

• https://feature4x.github.io/
• https://arxiv.org/abs/2503.20776

TANGO: Training-free Embodied AI Agents for Open-world Tasks

• https://openaccess.thecvf.com/content/CVPR2025/papers/Ziliotto_TANGO_Training-free_Embodied_AI_Agents_for_Open-world_Tasks_CVPR_2025_paper.pdf

Sketchtopia: A Dataset and Foundational Agents for Benchmarking Asynchronous Multimodal Communication with Iconic Feedback

• https://openaccess.thecvf.com/content/CVPR2025/papers/Khan_Sketchtopia_A_Dataset_and_Foundational_Agents_for_Benchmarking_Asynchronous_Multimodal_CVPR_2025_paper.pdf

SegAgent: Exploring Pixel Understanding Capabilities in MLLMs by Imitating Human Annotator Trajectories

• https://github.com/aim-uofa/SegAgent
• https://arxiv.org/abs/2503.08625

GUI-Xplore: Empowering Generalizable GUI Agents with One Exploration

• https://arxiv.org/html/2503.17709v1

RL-RC-DoT: A Block-level RL agent for Task-Aware Video Compression

• https://arxiv.org/html/2501.12216v2

SpiritSight Agent: Advanced GUI Agent with One Look

• https://arxiv.org/html/2503.03196v1

Do Visual Imaginations Improve Vision-and-Language Navigation Agents?

• https://www.akhilperincherry.com/VLN-Imagine-website/
• https://arxiv.org/html/2503.16394v1

Reasoning in visual navigation of end-to-end trained agents: a dynamical systems approach

• https://arxiv.org/abs/2503.08306

VADAR: Visual Agentic AI for Spatial Reasoning with a Dynamic API

• https://github.com/damianomarsili/VADAR
• https://arxiv.org/abs/2502.06787

Magma: A Foundation Model for Multimodal AI Agents

• https://microsoft.github.io/Magma/
• https://arxiv.org/abs/2502.13130

SketchAgent: Language-Driven Sequential Sketch Generation

• https://github.com/yael-vinker/SketchAgent
• https://arxiv.org/abs/2411.17673

Scalable Video-to-Dataset Generation for Cross-Platform Mobile Agents

• https://openaccess.thecvf.com/content/CVPR2025/papers/Jang_Scalable_Video-to-Dataset_Generation_for_Cross-Platform_Mobile_Agents_CVPR_2025_paper.pdf

From Multimodal LLMs to Generalist Embodied Agents: Methods and Lessons

• https://arxiv.org/abs/2412.08442

Optimus-2: Mulitimodal Minecraft Agent with Goal-Observation-Action Conditioned Policy

• https://cybertronagent.github.io/Optimus-2.github.io/
• https://arxiv.org/abs/2502.19902

Dual-Agent Optimization framework for Cross-Domain Few-Shot Segmentation

• https://openaccess.thecvf.com/content/CVPR2025/papers/Li_Dual-Agent_Optimization_framework_for_Cross-Domain_Few-Shot_Segmentation_CVPR_2025_paper.pdf

Taming Large Multimodal Agents for Ultra-low Bitrate Semantically Disentangled Image Compression

• https://arxiv.org/abs/2503.00399

Unlocking Video-LLM via Agent-of-Thoughts Distillation

• https://zhengrongz.github.io/AoTD/
• https://export.arxiv.org/abs/2412.01694

ShowUI: One Vision-Language-Action Model for GUI Visual Agent

• https://github.com/showlab/ShowUI
• https://openaccess.thecvf.com/content/CVPR2025/papers/Lin_ShowUI_One_Vision-Language-Action_Model_for_GUI_Visual_Agent_CVPR_2025_paper.pdf

多智能体：由多个智能体组成的协同系统，聚焦 "群体协作、冲突解决"，如自动驾驶车队、医疗多模态诊断团队。

多智能体 13

**MAGiC-SLAM: Multi-Agent Gaussian Globally Consistent SLAM

• https://arxiv.org/abs/2411.16785

MNE-SLAM: Multi-Agent Neural SLAM for Mobile Robots

• https://anonymous.4open.science/w/ins-BD50/
• https://openaccess.thecvf.com/content/CVPR2025/papers/Deng_MNE-SLAM_Multi-Agent_Neural_SLAM_for_Mobile_Robots_CVPR_2025_paper.pdf

Collaborative Tree Search for Enhancing Embodied Multi-Agent Collaboration

• https://openaccess.thecvf.com/content/CVPR2025/papers/Zu_Collaborative_Tree_Search_for_Enhancing_Embodied_Multi-Agent_Collaboration_CVPR_2025_paper.pdf

ComfyBench: Benchmarking LLM-based Agents in ComfyUI for Autonomously Designing Collaborative AI Systems

• https://github.com/xxyQwQ/ComfyBench
• https://arxiv.org/abs/2409.01392

NADER: Neural Architecture Design via Multi-Agent Collaboration

• https://arxiv.org/html/2412.19206v1#S1

V-Stylist: Video Stylization via Collaboration and Reflection of MLLM Agents

• https://zhengrongyue.github.io/v-stylist.github.io/
• https://arxiv.org/html/2503.12077v1

Unified Uncertainty-Aware Diffusion for Multi-Agent Trajectory Modeling

• https://arxiv.org/abs/2503.18589

MAC-Ego3D: Multi-Agent Gaussian Consensus for Real-Time Collaborative Ego-Motion and Photorealistic 3D Reconstruction

• https://github.com/Xiaohao-Xu/MAC-Ego3D
• https://arxiv.org/abs/2412.09723

MCCD: Multi-Agent Collaboration-based Compositional Diffusion for Complex Text-to-Image Generation

• https://openaccess.thecvf.com/content/CVPR2025/papers/Li_MCCD_Multi-Agent_Collaboration-based_Compositional_Diffusion_for_Complex_Text-to-Image_Generation_CVPR_2025_paper.pdf

Learning to Detect Objects from Multi-Agent LiDAR Scans without Manual Labels

• https://github.com/xmuqimingxia/DOtA
• https://arxiv.org/abs/2503.08421

TraF-Align: Trajectory-aware Feature Alignment for Asynchronous Multi-agent Perception

• https://github.com/zhyingS/TraF-Align
• https://arxiv.org/abs/2503.19391

Internet of Agents: Weaving a Web of Heterogeneous Agents for Collaborative Intelligence

• https://arxiv.org/abs/2407.07061
• https://github.com/openbmb/ioa

Synergistic Multi-Agent Framework with Trajectory Learning for Knowledge-Intensive Tasks

• https://arxiv.org/abs/2407.09893
• https://github.com/yueshengbin/SMART

Agent 技术基础与融合创新（按技术框架 / 融合范式划分）

这类是实现 Agent 的 "技术底座" 与 "创新方向"，覆盖从 "怎么建、怎么用、怎么评" 到 "技术交叉创新" 的全流程，是论文选题的核心方向。

大模型智能体（构建、应用、评估）：以大模型为核心的 Agent 技术体系，包含 "构建（如微调、推理增强）、应用（如工业控制、基层治理）、评估（如任务成功率、伦理对齐）" 全链路，是当前主流技术基础。

大模型智能体

Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models

• https://arxiv.org/abs/2408.08926
• https://github.com/andyzorigin/cyber-bench
• https://github.com/andyzorigin/cybench

REGENT: A Retrieval-Augmented Generalist Agent That Can Act In-Context in New Environments

• https://arxiv.org/abs/2412.04759

SPA-BENCH: A COMPREHENSIVE BENCHMARK FOR SMARTPHONE AGENT EVALUATION

• https://openreview.net/forum?id=ikXjMk8RUs
• https://github.com/ai-agents-2030/SPA-Bench

SWEET-RL: Training Multi-Turn LLM Agents on Collaborative Reasoning Tasks

• https://arxiv.org/abs/2503.15478
• https://github.com/facebookresearch/sweet_rl

Graph+AI Agents：将 "图技术（知识图谱、任务依赖图等）" 与 Agent 融合的创新范式，聚焦 "提升推理效率、优化记忆管理、增强多体协同"，如基于知识图谱的多跳问答 Agent、图驱动的工具调用优化。

Graph+AI Agents

AFlow: Automating Agentic Workflow Generation

• https://arxiv.org/abs/2410.10762
• https://github.com/geekan/metagpt
• https://github.com/foundationagents/aflow

Scaling Large Language Model-based Multi-Agent Collaboration

• https://arxiv.org/abs/2406.07155
• https://github.com/openbmb/chatdev

Agent+RL

AgentGym-RL: An Open-Source Framework to Train LLM Agents for Long-Horizon Decision Making via Multi-Turn RL

• https://openreview.net/forum?id=ZgCCDwcGwn
• https://github.com/WooooDyy/AgentGym-RL
• https://huggingface.co/datasets/AgentGym/AgentGym-RL-Data-ID
• https://agentgym-rl.github.io/

From Reasoning to Agentic: Credit Assignment in Reinforcement Learning for Large Language Models

• https://arxiv.org/abs/2604.09459
• https://github.com/xxzcc/Awesome-Credit-Assignment-in-LLM-RL

ArenaRL: Scaling RL for Open-Ended Agents via Tournament-based Relative Ranking

• https://arxiv.org/pdf/2601.06487

Agent-R1: Training Powerful LLM Agents with End-to-End Reinforcement Learning

• https://arxiv.org/pdf/2511.14460
• https://github.com/AgentR1/Agent-R1

AGENTGYM-RL: AN OPEN-SOURCE FRAMEWORK TO TRAIN LLM AGENTS FOR LONG-HORIZON DECISION MAKING VIA MULTI-TURN RL

• https://arxiv.org/pdf/2509.08755
• https://github.com/WooooDyy/AgentGym-RL

SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning

• arXiv:2602.08234
• https://github.com/aiming-lab/SkillRL

ProRL Agent: Rollout-as-a-Service for RL Training of Multi-Turn LLM Agents

• https://arxiv.org/abs/2603.18815
• https://github.com/NVIDIA-NeMo/ProRL-Agent-Server

AgentV-RL: Scaling Reward Modeling with Agentic Verifier

• https://arxiv.org/abs/2604.16004
• https://github.com/JiazhengZhang/AgentV-RL

Relax: An Asynchronous Reinforcement Learning Engine for Omni-Modal Post-Training at Scale

• https://arxiv.org/abs/2604.11554
• https://github.com/redai-infra/Relax?utm_source=chatgpt.com

UDAAgent: Large-Scale Agentic RL for High-Performance CUDA Kernel Generation

• https://arxiv.org/abs/2602.24286
• https://cuda-agent.github.io/

MemPO: Self-Memory Policy Optimization for Long-Horizon Agents

• https://arxiv.org/abs/2603.00680
• https://github.com/TheNewBeeKing/MemPO
• https://huggingface.co/collections/NewBeeKing/mempo

Exploring Reasoning Reward Model for Agents

• https://arxiv.org/pdf/2601.22154
• https://github.com/kxfan2002/Reagent

Agent memory

PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents- https://arxiv.org/abs/2603.03296- https://github.com/TIMAN-group/PlugMem

Lightweight LLM Agent Memory with Small Language Models

• https://arxiv.org/pdf/2604.07798

FinMem: A Performance-Enhanced LLM Trading Agent with Layered Memory and Character Design

• https://ojs.aaai.org/index.php/AAAI-SS/article/view/31290
• https://github.com/pipiku915/FinMem-LLM-StockTrading这篇文章的核心是研究如何通过设计一个具有分层记忆和角色设定的LLM代理框架来增强金融交易决策能力。它属于Agent-Memory（代理记忆）的研究范畴，特别是针对金融领域的应用。

Optimus-1 : Hybrid Multimodal Memory Empowered Agents Excel in Long-Horizon Tasks

• https://proceedings.neurips.cc/paper_files/paper/2024/file/5949a8750a110ce1f0631b1776c500a2-Paper-Conference.pdf
• https://github.com/JiuTian-VL/Optimus-1这篇文章的核心是研究如何通过设计一个混合多模态记忆模块来增强代理在复杂任务中的表现。这种记忆模块不仅提供了必要的世界知识，还通过总结和存储多模态经验来指导代理的决策过程。这种设计显著提高了代理在长时域任务中的成功率和效率，是Agent-Memory研究的一个重要进展。

Mobile-Agent-v2: Mobile Device Operation Assistant with Effective Navigation via Multi-Agent Collaboration

• https://proceedings.neurips.cc/paper_files/paper/2024/file/0520537ba799d375b8ff5523295c337a-Paper-Conference.pdf
• https://github.com/X-PLUG/MobileAgent

AGENTPOISON: Red-teaming LLM Agents via Poisoning Memory or Knowledge Bases

• https://proceedings.neurips.cc/paper_files/paper/2024/file/eb113910e9c3f6242541c1652e30dfd6-Paper-Conference.pdf
• https://github.com/AI-secure/AgentPoison

Memory in the Age of Al Agents: A Survey

• https://arxiv.org/pdf/2512.13564
• https://github.com/Shichun-Liu/Agent-Memory-Paper-List

Agentic Memory: Learning Unifed Long-Term and Short-Term Memory Management for Large Language Model Agents

• https://arxiv.org/pdf/2601.01885

A-Mem: Agentic Memory for LLM Agents

• https://arxiv.org/pdf/2502.12110
• https://github.com/WujiangXu/A-mem
• https://github.com/WujiangXu/A-mem-sysA-MEM 通过动态记忆组织和演化机制，在多跳推理任务中取得了显著的性能提升，F1 分数和 BLEU-1 分数分别比现有基线模型提升了 20% 以上。同时，A-MEM 在保持高质量回答的同时，显著减少了回答所需的 token 数量，提高了计算效率。这些结果表明，A-MEM 为 LLM 代理提供了一种高效、灵活且适应性强的记忆管理解决方案。

SRMT: SHARED MEMORY FOR MULTI-AGENT LIFE LONG PATHFINDING

• https://arxiv.org/pdf/2501.13200
• https://github.com/Aloriosa/srmtSRMT 通过引入共享记忆机制，显著提升了多智能体系统在路径规划任务中的协调能力和泛化能力。它不仅在复杂环境中表现出色，还能够在训练时未见过的环境中保持高效。这项研究为多智能体系统中的记忆共享和路径规划提供了一种新的解决方案，展示了其在去中心化多智能体系统中的潜力。

LLM-Empowered Embodied Agent for Memory-Augmented Task Planning in Household Robotics

• https://arxiv.org/pdf/2504.21716
• https://github.com/marc1198/chat-hsr论文提出了一种基于 LLM 的家庭机器人系统，通过记忆增强的任务规划和模块化的代理协作，显著提升了机器人在动态家庭环境中的任务规划能力和长期记忆管理能力。Qwen2.5 在任务规划和知识库任务中表现最佳，而 LLaMA3.1 在路由任务中表现优异。这项研究展示了 LLM 在家庭机器人任务规划和记忆管理中的潜力，并为未来的研究提供了新的方向。

STMA: A Spatio-Temporal Memory Agent for Long-Horizon Embodied Task Planning

• https://arxiv.org/pdf/2502.10177
• https://github.com/SP4595/STMA-A-Spatio-Temporal-Memory-Agent-for-Long-Horizon-Embodied-Task-Planning/blob/main/eb-example.pdfSTMA 通过整合时空记忆和规划者-批评者机制，显著提升了代理在动态环境中的任务规划和执行能力。实验结果表明，STMA 在成功率和平均得分上均优于现有最先进模型，尤其是在复杂任务中表现更为出色。此外，STMA 在使用开源模型时也表现出色，验证了其在实际应用中的潜力。STMA 在任务成功率上比最佳基线模型提高了 31.25%。

VideoAgent: A Memory-augmented Multimodal Agent for Video Understanding

• https://arxiv.org/pdf/2403.11481
• https://github.com/YueFan1014/VideoAgentVideoAgent 通过引入统一的记忆机制和多模态工具使用，有效地解决了长视频理解中的挑战，展示了其在复杂视频任务中的优势。这项研究为未来的视频理解研究提供了新的方向，特别是在如何更好地整合记忆和多模态工具方面。

HIAGENT: Hierarchical Working Memory Management for Solving Long-Horizon Agent Tasks with Large Language Model

• https://arxiv.org/pdf/2408.09559
• https://github.com/HiAgent2024/HiAgentHIAGENT 通过层次化的记忆管理机制，显著提高了基于 LLM 的代理在长期任务中的表现。实验结果表明，HIAGENT 在成功率、进步率、平均步数、上下文效率和运行时间等多个指标上均优于现有的标准方法。此外，HIAGENT 的设计允许其灵活地集成到其他代理框架中，为未来的研究提供了新的方向。

MIRIX: Multi-Agent Memory System for LLM-Based Agents

• https://arxiv.org/pdf/2507.07957
• https://github.com/Mirix-AI/MIRIXMIRIX 通过其结构化的记忆组件和多智能体框架，显著提高了基于 LLM 的代理在长期任务中的表现。实验结果表明，MIRIX 在多模态输入处理和长期对话基准测试中均优于现有记忆系统。此外，MIRIX 的设计使其能够灵活地集成到其他代理框架中，为未来的研究和应用提供了新的方向。

Agent 学习支撑资源（按学习工具 / 路径划分）

这类是入门与论文写作的 "辅助工具包"，为前两类内容提供学习方法、资料与工具支持，解决 "怎么学、怎么上手、怎么查资料" 的问题。

Agent必备基础理论和开发框架

Agent 开发必读：包含经典论文（如 Transformer、PaLM-E 相关）、开发框架（LangChain、CrewAI）、学习路径（入门→进阶→研究）、论文写作建议，是支撑前两类知识学习与实践的资源集合。

推荐学习顺序：基础架构→训练优化→应用与Agent

基础架构与模型设计

Attention Is All You Need

• https://proceedings.neurips.cc/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf
• https://github.com/tensorflow/tensor2tensor

Language Models are Few-Shot Learners

• https://proceedings.neurips.cc/paper_files/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf
• https://github.com/openai/gpt-3

GShard: Scaling Giant Models with Conditional Computation

• https://arxiv.org/pdf/2006.16668

Switch Transformers: Scaling to Trillion Parameter Models

• https://www.jmlr.org/papers/volume23/21-0998/21-0998.pdf
• https://github.com/tensorflow/mesh

训练优化与扩展

Training Compute-Optimal Large Language Models

• https://arxiv.org/pdf/2203.15556
• https://github.com/karpathy/llama2.c

ZeRO: Memory Optimizations Toward Training Trillion Parameter Models

• https://arxiv.org/pdf/1910.02054
• https://github.com/microsoft/DeepSpeed

Mixed Precision Training

• https://arxiv.org/pdf/1710.03740

LoRA: Low-Rank Adaptation of Large Language Models

• https://arxiv.org/pdf/2106.09685v1/1000
• https://github.com/microsoft/LoRA

多模态与跨领域

Learning Transferable Visual Models From Natural Language Supervision

• http://proceedings.mlr.press/v139/radford21a/radford21a.pdf
• https://github.com/openai/CLIP

Flamingo: a Visual Language Model for Few-Shot Learning

• https://proceedings.neurips.cc/paper_files/paper/2022/file/960a172bc7fbf0177ccccbb411a7d800-Paper-Conference.pdf

PaLM-E: An Embodied Multimodal Language Model

• https://openreview.net/pdf?id=VTpHpqM3Cf

推理与知识增强

Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

• https://proceedings.neurips.cc/paper_files/paper/2022/file/9d5609613524ecf4f15af0f7b31abca4-Paper-Conference.pdf
• https://github.com/microsoft/guidance

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

• https://proceedings.neurips.cc/paper/2020/file/6b493230205f780e1bc26945df7481e5-Paper.pdf
• https://github.com/huggingface/transformers

Toolformer: Language Models Can Teach Themselves to Use Tools

• https://proceedings.neurips.cc/paper_files/paper/2023/file/d842425e4bf79ba039352da0f658a906-Paper-Conference.pdf
• https://github.com/lucidrains/toolformer-pytorch

安全与对齐

Training Language Models to Follow Instructions with Human Feedback

• https://proceedings.neurips.cc/paper_files/paper/2022/file/b1efde53be364a73914f58805a001731-Paper-Conference.pdf
• https://github.com/openai/following-instructions-human-feedback

Red Teaming Language Models with Language Models

• https://arxiv.org/pdf/2202.03286
• https://github.com/google-research/google-research

Constitutional AI: Harmlessness from AI Feedback

• https://ai-plans.com/file_storage/4f32fa39-3a01-46c7-878e-c92b7aa7165f_2212.08073v1.pdf
• https://github.com/anthropics/constitutionalharmlessnesspaper

应用与Agent系统

ReAct: Synergizing Reasoning and Acting in Language Models

• https://par.nsf.gov/servlets/purl/10451467
• https://github.com/ysymyth/ReAct

Voyager: An Open-Ended Embodied Agent with Large Language Models

• https://arxiv.org/pdf/2305.16291
• https://github.com/MineDojo/Voyager

Agent经典论文

推理基础

是地基

Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

• https://arxiv.org/pdf/2201.11903

Self-Consistency Improves Chain of Thought Reasoning in Language Models

• https://arxiv.org/pdf/2203.11171

Tree of Thoughts: Deliberate Problem Solving with Large Language Models

• https://proceedings.neurips.cc/paper_files/paper/2023/file/271db9922b8d1f4dd7aaef84ed5ac703-Paper-Conference.pdf
• https://github.com/princeton-nlp/tree-of-thought-llm

Self-Refine: Iterative Refinement with Self-Feedback

• https://arxiv.org/pdf/2303.17651
• https://github.com/madaan/self-refine

规划

推理是"想清楚"，规划是"排好步骤"

Language Models as Zero-Shot Planners: Extracting Actionable Knowledge for Embodied Agents

• https://arxiv.org/pdf/2201.07207
• https://github.com/huangwl18/language-planner

Plan-and-Solve Prompting: Improving Zero-Shot Chain-of-Thought Reasoning by Large Language Models

• https://arxiv.org/pdf/2305.04091
• https://github.com/AGI-Edgerunners/Plan-and-Solve-Prompting

Reasoning with Language Model is Planning with World Model

• https://arxiv.org/pdf/2305.14992
• https://github.com/Ber666/llm-reasoners

推理+行动

ReAct: Synergizing Reasoning and Acting in Language Models

• https://research.google/blog/react-synergizing-reasoning-and-acting-in-language-models/

Reflexion: Language Agents with Verbal Reinforcement Learning

• https://arxiv.org/pdf/2303.11366
• https://github.com/noahshinn024/reflexion

Language Agent Tree Search Unifies Reasoning, Acting, and Planning in Language Models

• https://arxiv.org/pdf/2310.04406
• https://github.com/lapisrocks/LanguageAgentTreeSearch

工具使用

Toolformer: Language Models Can Teach Themselves to Use Tools

• https://arxiv.org/pdf/2302.04761

Gorilla: Large Language Model Connected with Massive APIs

• https://arxiv.org/pdf/2305.15334
• https://github.com/ShishirPatil/gorilla

ToolGen: Unified Tool Retrieval and Calling via Generation

• https://github.com/Reason-Wang/ToolGen
• https://arxiv.org/pdf/2410.03439

记忆

Generative Agents: Interactive Simulacra of Human Behavior

• https://arxiv.org/pdf/2304.03442
• https://github.com/joonspk-research/generative_agents

MemGPT: Towards LLMs as Operating Systems

• https://arxiv.org/pdf/2310.08560
• https://github.com/letta-ai/letta

MemO: Building Production-Ready AI Agents with Scalable Long-Term Memory

• https://arxiv.org/pdf/2504.19413
• https://mem0.ai/research

MemoryBank: Enhancing Large Language Models with Long-Term Memory

• https://arxiv.org/pdf/2305.10250
• https://github.com/zhongwanjun/MemoryBank-SiliconFriend

多智能体

CAMEL: Communicative Agents for "Mind" Exploration of Large Language Model Society

• https://proceedings.neurips.cc/paper_files/paper/2023/file/a3621ee907def47c1b952ade25c67698-Paper-Conference.pdf
• https://github.com/camel-ai/camel

AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation

• https://arxiv.org/pdf/2308.08155
• https://github.com/microsoft/autogen

MetaGPT: Meta Programming for a Multi-Agent Collaborative Framework

• https://arxiv.org/pdf/2308.00352
• https://github.com/FoundationAgents/MetaGPT

技能

Agent Skills for Large Language Models: Architecture, Acquisition, Security, and the Path Forward

• https://arxiv.org/pdf/2602.12430
• https://github.com/scienceaix/agentskills

Agentic RL

PPO: InstructGPT: Training Language Models to Follow Instructions with Human Feedback

• https://arxiv.org/pdf/2203.02155

DPO: Direct Preference Optimization - Your Language Model is Secretly a Reward Model

• https://arxiv.org/pdf/2305.18290

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

• https://arxiv.org/pdf/2501.12948

Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning

• https://arxiv.org/pdf/2503.09516
• https://github.com/PeterGriffinJin/Search-R1

AgentRL: Scaling Agentic Reinforcement Learning with a Multi-Turn, Multi-Task Framework

• https://arxiv.org/pdf/2510.04206
• https://github.com/THUDM/AgentRL

项目资源

谷歌321个Agent落地案例，全球顶级企业，6大核心场景

• https://blog.google/products/google-cloud/gen-ai-business-use-cases/#customer-agents

500个开源的AI Agent智能体

• https://github.com/ashishpatel26/500-AI-Agents-Projects?tab=readme-ov-file500个开源的智能体应用，涵盖健康，金融，教育，零售各个行业，不同的开发框架也做了分类