Awesome Agent Swarm

Multi-agent swarm systems, orchestration frameworks, swarm intelligence, agent communication protocols, and collaborative AI.

Contents

• Taxonomy
• Swarm Frameworks
• Orchestration and Workflow
• Agent Communication and Protocols
• Swarm Intelligence
• Role-Based Agent Teams
• Task Decomposition and Planning
• Swarm Coding and Engineering
• Safety and Governance
• Key Research Papers
• Benchmarks and Evaluation
• Community and Resources

Taxonomy

graph LR
    Root["Agent Swarm"] --> Core["Core<br/>Frameworks"]
    Root --> Coordination["Coordination<br/>& Communication"]
    Root --> Patterns["Swarm<br/>Patterns"]
    Root --> Applications["Applications<br/>& Safety"]

    Core --> Frameworks["Swarm<br/>Frameworks"]
    Core --> Orchestration["Orchestration<br/>& Workflow"]

    Coordination --> Protocols["A2A & MCP<br/>Protocols"]
    Coordination --> TaskDecomp["Task Decomposition<br/>& Planning"]

    Patterns --> Intelligence["Swarm<br/>Intelligence"]
    Patterns --> RoleTeams["Role-Based<br/>Agent Teams"]

    Applications --> Coding["Swarm Coding<br/>& Engineering"]
    Applications --> Safety["Safety &<br/>Governance"]

Loading

Swarm Frameworks

Core frameworks for building and managing multi-agent swarm systems.

• AutoGen - Programming framework for agentic AI by Microsoft. Build multi-agent applications with conversational patterns and group chat. by @microsoft (57,099 stars)
• AgentScope - Production-ready multi-agent framework with ReAct, memory, planning, and A2A support. Build and run agents you can see, understand and trust. by @agentscope-ai (23,752 stars)
• Mastra - TypeScript framework for building AI-powered multi-agent applications. From the team behind Gatsby. by @mastra-ai (23,012 stars)
• Swarm (OpenAI) - Educational framework exploring lightweight multi-agent orchestration. Demonstrates handoffs and routines patterns for agent coordination. by @OpenAI (21,311 stars)
• OpenAI Agents Python - Production-ready multi-agent framework from OpenAI. Features agent handoffs, guardrails, and tracing for swarm workflows. by @OpenAI (20,784 stars)
• Google ADK - Open-source Python toolkit by Google for building, evaluating, and deploying multi-agent systems with orchestration support. by @google (18,988 stars)
• Microsoft Agent Framework - Framework for building, orchestrating and deploying multi-agent systems with support for Python and .NET. by @microsoft (9,462 stars)
• Spring AI Alibaba - Enterprise-grade multi-agent framework for Java developers by Alibaba. Spring ecosystem integration with agent orchestration. by @alibaba (9,243 stars)
• PraisonAI - Low-code multi-agent framework with 100+ built-in tools. Define agent swarms via YAML configuration. by @MervinPraison (6,934 stars)
• Swarms - Enterprise-grade multi-agent orchestration framework. Sequential, parallel, hierarchical, and mesh swarm topologies. by @kyegomez (6,235 stars)
• Agency Swarm - Multi-agent orchestration framework built on OpenAI Agents SDK. Define agent teams with customizable roles and communication flows. by @VRSEN (4,212 stars)
• EvoMap - Agent Swarm platform with task decomposition, Worker Pool orchestration, Evolution Circles, AI Council multi-agent governance, Privacy Computing, and ARC-AGI-2 arena. by @EvoMap (1,898 stars)

Orchestration and Workflow

Orchestration engines, workflow builders, and pipeline frameworks for coordinating agent swarms.

• Dify - Production-ready platform for agentic workflow development. Visual workflow builder with multi-agent orchestration, RAG pipeline, and model management. by @langgenius (137,817 stars)
• DeerFlow - Open-source long-horizon SuperAgent harness by ByteDance. Multi-agent collaboration for research, coding, and content creation. by @bytedance (61,643 stars)
• LangGraph - Build resilient language agents as graphs. Low-level orchestration framework for stateful, multi-actor applications with durable execution. by @langchain-ai (29,306 stars)
• Agent Squad - AWS framework for managing multiple AI agents and handling complex conversations with intelligent routing. by @awslabs (7,570 stars)

Agent Communication and Protocols

Standards and protocols for inter-agent messaging, discovery, and interoperability.

• Google A2A - Google's open Agent-to-Agent protocol. Enables agent discovery, secure collaboration, and long-running tasks while preserving agent opacity. by @google (23,201 stars)
• Coral Anemoi - Semi-centralized multi-agent coordination via Agent-to-Agent Communication MCP server. Enables cross-framework agent collaboration. by @Coral-Protocol (374 stars)
• GEP MCP Server - MCP Server for Genome Evolution Protocol. Exposes swarm evolution tools to Claude Desktop, Cursor, and any MCP client. by @EvoMap (1 stars)

Swarm Intelligence

Emergent behavior, collective reasoning, and self-organizing multi-agent systems.

• OWL - Optimized Workforce Learning framework built on CAMEL-AI. #1 on GAIA benchmark (69.09) among open-source multi-agent systems for real-world task automation. by @camel-ai (19,654 stars)
• CAMEL - The first multi-agent framework. Finding the Scaling Law of Agents through role-playing and communicative agent collaboration. by @camel-ai (16,693 stars)
• ClawTeam - Agent Swarm Intelligence framework. Agents self-organize into collaborative teams with dynamic task allocation, inter-agent messaging, and git worktree isolation. by @HKUDS (4,769 stars)
• LatentMAS - Latent collaboration in multi-agent systems. Agents reason and collaborate in continuous latent space instead of natural language, reducing communication overhead. by @Gen-Verse (873 stars)

Role-Based Agent Teams

Frameworks that organize agents into specialized roles for collaborative task execution.

• MetaGPT - Virtual software company via multi-agent collaboration. SOPs encoded as prompts assign PM, architect, developer, and QA roles. by @FoundationAgents (67,082 stars)
• CrewAI - Framework for orchestrating role-playing, autonomous AI agents. Define crews with specialized roles, goals, and backstories for collaborative tasks. by @crewAIInc (48,922 stars)
• ChatDev - Virtual software company via LLM-powered multi-agent conversation chains. Agents play CEO, CTO, programmer, and tester roles. by @OpenBMB (32,715 stars)
• HiClaw - Collaborative Multi-Agent OS with Manager-Workers architecture. Human-in-the-loop task coordination with enterprise-grade security. by @agentscope-ai (4,104 stars)

Task Decomposition and Planning

Systems for breaking complex goals into subtasks, building execution DAGs, and coordinating parallel agent work.

• MindSearch - Multi-agent web search engine. Decomposes search queries into sub-tasks, delegates to specialized agents, and aggregates results. by @InternLM (6,839 stars)
• Open Multi-Agent - TypeScript multi-agent orchestration via single runTeam() call. Auto-decomposes goals into task DAGs and runs agents in parallel. by @JackChen-me (5,708 stars)
• AFlow - Automated multi-agent workflow generation via Monte Carlo tree search. Designs optimal agent topologies for given tasks. by @geekan (1,500 stars)

Swarm Coding and Engineering

Agent swarms applied to collaborative software development and engineering workflows.

No projects yet. Submit one!

Safety and Governance

Guardrails, policy engines, and governance frameworks for multi-agent systems.

• NeMo Guardrails - NVIDIA's toolkit for adding programmable guardrails to LLM systems. Policy-based safety controls for multi-agent deployments. by @NVIDIA (5,977 stars)

Key Research Papers

Surveys

• LLMs Working in Harmony: A Survey on Building Effective LLM-Based Multi Agent Systems (arXiv'25) - Covers architecture, memory, planning, and frameworks for multi-agent LLM systems.
• Multi-Agent Collaboration Mechanisms: A Survey of LLMs (arXiv'25) - Characterizes collaboration by actors, types (cooperation, competition, coopetition), structures, and protocols.
• A Comprehensive Survey of Self-Evolving AI Agents (arXiv'25) - Taxonomy of single-agent, multi-agent, and domain-specific evolution.
• Large Language Model based Multi-Agents: A Survey of Progress and Challenges (IJCAI'24) - Agent profiling, communication methods, and skill development.

Swarm Orchestration and Architecture

• AgentOrchestra: A Hierarchical Multi-Agent Framework for General-Purpose Task Solving (arXiv'25) - Conductor-inspired hierarchical framework with MCP Manager Agent. 83.39% on GAIA benchmark.
• MAS-Orchestra: Understanding Multi-Agent Reasoning Through Holistic Orchestration (arXiv'26) - Multi-agent orchestration as a function-calling RL problem. Introduces MASBENCH along 5 dimensions.
• AdaptOrch: Task-Adaptive Orchestration (arXiv'26) - Dynamic topology selection (parallel, sequential, hierarchical, hybrid) based on task dependency graphs. 12-23% over static baselines.
• Puppeteer: Evolving Orchestration via Reinforcement Learning (arXiv'25) - Centralized orchestrator trained via RL to dynamically direct agents in response to evolving task states.

Swarm Intelligence and Emergence

• SwarmSys: Decentralized Swarm-Inspired Agents for Scalable Reasoning (arXiv'25) - Explorers, Workers, and Validators with pheromone-inspired reinforcement. Coordination scaling vs model scaling.
• SIER: Swarm Intelligence Enhancing Reasoning (arXiv'25) - Kernel density estimation and non-dominated sorting for swarm-guided LLM reasoning.
• Multi-Agent Systems Powered by LLMs: Applications in Swarm Intelligence (arXiv'25) - LLMs integrated into multi-agent simulations for ant colony foraging and bird flocking.
• LatentMAS: Latent Collaboration in Multi-Agent Systems (arXiv'25) - Agents collaborate in continuous latent space instead of natural language.
• GPTSwarm: Language Agents as Optimizable Graphs (ICML'24) - Graph-based optimization of agent collaboration topologies.
• Scaling Large-Language-Model-based Multi-Agent Collaboration (arXiv'24) - Scaling laws and topology analysis for multi-agent collaboration.

Multi-Agent Collaboration and Evolution

• Self-Evolving Multi-Agent Collaboration Networks (ICLR'25) - Multi-agent systems that evolve their collaboration patterns through experience.
• AFlow: Automating Agentic Workflow Generation (ICLR'25) - Monte Carlo tree search for automated multi-agent workflow design.
• Automated Design of Agentic Systems (ADAS) (ICLR'25) - Meta-learning for automatic agent system design.
• AgentVerse: Facilitating Multi-Agent Collaboration (ICLR'24) - Emergent behaviors in multi-agent environments.
• SEMAG: Self-Evolutionary Multi-Agent Code Generation (arXiv'26) - Self-evolutionary agents that auto-upgrade backbone models.
• SAGE: Multi-Agent Self-Evolution for LLM Reasoning (arXiv'26) - Four co-evolving agents from shared LLM backbone.
• Group-Evolving Agents (arXiv'26) - Agent groups as evolutionary units with experience sharing. 71.0% on SWE-bench Verified.

Role-Based Teams and Software Engineering

• MetaGPT: Meta Programming for Multi-Agent Collaboration (ICLR'24) - SOPs encoded as prompts for multi-agent software development.
• Communicative Agents for Software Development (ACL'24) - ChatDev: virtual software company through multi-agent conversation chains.
• Agyn: Team-Based Autonomous Software Engineering (arXiv'26) - Coordination, research, implementation, and review agents. 72.2% on SWE-bench 500.
• Self-Organizing Multi-Agent Systems for Continuous Software Development (arXiv'26) - Manager agents dynamically hire/assign/fire workers. Multi-day continuous development.
• AgentMesh: Cooperative Multi-Agent Framework for Software Development (arXiv'25) - Planner, Coder, Debugger, Reviewer agents for end-to-end development.
• MoRAgent: Parameter Efficient Agent Tuning with Mixture-of-Roles (arXiv'25) - Reasoner, executor, and summarizer roles via specialized LoRA groups.

Task Decomposition and Planning

• Modular Task Decomposition and Dynamic Collaboration in Multi-Agent Systems (arXiv'25) - Hierarchical sub-task decomposition with dynamic scheduling and constraint parsing.
• CodeAgents: Token-Efficient Multi-Agent Reasoning (arXiv'25) - Multi-agent reasoning codified into modular pseudocode. 55-87% token reduction.
• ReWOO: Decoupling Reasoning from Observations (ICML'24) - Efficient tool-augmented reasoning by decoupling planning from execution.
• Tree of Thoughts: Deliberate Problem Solving with LLMs (NeurIPS'23) - Tree-structured reasoning for complex problem solving.
• HuggingGPT: Solving AI Tasks with ChatGPT and its Friends (NeurIPS'23) - Task decomposition across specialized AI models.

Agent Communication Protocols

• Agent Communication Protocol (ACP) (arXiv'26) - Standardized A2A framework with federated orchestration, semantic intent mapping, and zero-trust security. 40% latency reduction.
• AgentMaster: Multi-Agent Framework Using A2A and MCP (EMNLP'25) - Combined A2A and MCP protocols for multimodal information retrieval. BERTScore F1 96.3%.
• Critical Analysis of A2A and MCP Integration for Scalable Agent Systems (arXiv'25) - Semantic interoperability, security, and governance challenges in A2A+MCP integration.

Safety, Governance, and Alignment

• AGENTSAFE: Unified Framework for Ethical Assurance and Governance in Agentic AI (arXiv'25) - Design, runtime, and audit controls covering the agentic loop (plan, act, observe, reflect).
• MI9: Integrated Runtime Governance Framework for Agentic AI (arXiv'25) - Agency-risk index, semantic telemetry, conformance engines, and graduated containment.
• The Alignment Flywheel: Governance-Centric Hybrid Multi-Agent Architecture (arXiv'26) - Proposer + Safety Oracle with patch locality for runtime governance.
• OpenGuardrails: Context-Aware AI Guardrails Platform (arXiv'25) - Context-aware safety detection and model-manipulation prevention.

Benchmarks and Evaluation

• MultiAgentBench (arXiv'25) - Evaluates collaboration and competition across star, chain, tree, and graph topologies.
• SwarmBench: Benchmarking LLMs' Swarm Intelligence (arXiv'25) - Pursuit, Synchronization, Foraging, Flocking, Transport tasks under decentralized constraints.
• Silo-Bench: Evaluating Distributed Coordination in Multi-Agent LLM Systems (arXiv'26) - 30 algorithmic tasks across 3 communication complexity levels. Reveals the "Communication-Reasoning Gap".
• Collab-Overcooked: Evaluating Fine-Grained Collaboration (EMNLP'25) - Competence boundary awareness, communication, and dynamic adaptation assessment.
• MASBENCH: Controlled Benchmark for Multi-Agent Orchestration (arXiv'26) - Tasks characterized along Depth, Horizon, Breadth, Parallel, and Robustness dimensions.
• SWE-bench (ICLR'24) - Can agent swarms resolve real-world GitHub issues?
• AgentBench (ICLR'24) - Multi-dimensional evaluation of LLMs as agents.
• GAIA (ICLR'23) - General AI assistant capabilities benchmark.

Community and Resources

No projects yet. Submit one!

Footnotes

Maintained by EvoMap. See contributing guidelines for how to submit a project or paper.

Also check out Awesome Agent Evolution for agent self-evolution, memory systems, and autonomous self-improvement.