OpenSpiel is a research framework that provides a wide range of environments (from simple matrix games to complex board games such as Chess, Go, and Poker) and implements various reinforcement learning and search algorithms (e.g., value iteration, policy gradient methods, MCTS). Its modular C++ core and Python bindings allow users to plug in custom algorithms, define new games, and compare performance across standard benchmarks. Designed for extensibility, it supports single and multi-agent settings, enabling study of cooperative and competitive scenarios. Researchers leverage OpenSpiel to prototype algorithms quickly, run large-scale experiments, and share reproducible code.

Reinforcement Learning Agents for PettingZoo Games is a Python-based code library delivering off-the-shelf DQN, PPO, and A2C algorithms for multi-agent reinforcement learning on PettingZoo environments. It features standardized training and evaluation scripts, configurable hyperparameters, integrated TensorBoard logging, and support for both competitive and cooperative games. Researchers and developers can clone the repo, adjust environment and algorithm parameters, run training sessions, and visualize metrics to benchmark and iterate quickly on their multi-agent RL experiments.

RoboCup Rescue Agent Simulation is an open-source framework that models urban disaster environments where multiple AI-driven agents collaborate to locate and rescue victims. It offers interfaces for navigation, mapping, communication, and sensor integration. Users can script custom agent strategies, run batch experiments, and visualize agent performance metrics. The platform supports scenario configuration, logging, and result analysis to accelerate research in multi-agent systems and disaster response algorithms.

ROCKET-1 is an open-source AI agent orchestration platform designed for building advanced multi-agent systems. It lets users define agent pipelines using a modular API, enabling seamless chaining of language models, plugins, and data stores. Core features include semantic memory to maintain context across sessions, dynamic tool integration for external APIs and databases, and built-in monitoring dashboards to track performance metrics. Developers can customize workflows with minimal code, scale horizontally via containerized deployments, and extend functionality through a plugin architecture. ROCKET-1 supports real-time debugging, automated retries, and security controls, making it ideal for customer support bots, research assistants, and enterprise automation tasks.

VacuumWorld is an open-source simulation platform designed to facilitate the development and evaluation of multi-agent reinforcement learning algorithms. It provides grid-based environments where virtual vacuum cleaner agents operate to detect and remove dirt patches across customizable layouts. Users can adjust parameters such as grid size, dirt distribution, stochastic movement noise, and reward structures to model diverse scenarios. The framework includes built-in support for agent communication protocols, real-time visualization dashboards, and logging utilities for performance tracking. With simple Python APIs, researchers can quickly integrate their RL algorithms, compare cooperative or competitive strategies, and conduct reproducible experiments, making VacuumWorld ideal for academic research and teaching.

Salieri is an innovative platform designed to streamline AI project development through multi-agent applications. By leveraging advanced AI technologies, Salieri enhances productivity and efficiency, making it easier for teams to automate workflows. Salieri's intuitive design and powerful functionalities allow users to translate detailed ideas into interactive, illustrated stories, perfect for narrative-driven projects, games, and more. Offering robust and efficient systems, Salieri integrates knowledge graphs and formal engines to improve the accuracy and cost-effectiveness of AI models.

SARL isms for decision-making and supports the dynamic with the Eclipse IDE, offering editor support, code generation, debugging, and testing tools. The runtime engine can target various platforms, including simulation frameworks (e.g., MadKit, Janus) and real-world systems in robotics and IoT. Developers can structure complex MAS applications by assembling modular skills and protocols, simplifying the development of adaptive, distributed AI systems.

Shepherding is an open-source simulation framework designed for reinforcement learning researchers and developers to study and implement multi-agent herding tasks. It provides a Gym-compatible environment where agents can be trained to perform behaviors such as flanking, collecting, and dispersing target groups across continuous or discrete spaces. The framework includes modular reward shaping functions, environment parameterization, and logging utilities for monitoring training performance. Users can define obstacles, dynamic agent populations, and custom policies using TensorFlow or PyTorch. Visualization scripts generate trajectory plots and video recordings of agent interactions. Shepherding’s modular design allows seamless integration with existing RL libraries, enabling reproducible experiments, benchmarking of novel coordination strategies, and rapid prototyping of AI-driven herding solutions.

OpenAI Swarm is a versatile orchestration library enabling parallel execution and consensus-driven decision-making across multiple AI agents. It broadcasts tasks to independent model instances, aggregates their outputs, and applies configurable voting or ranking schemes to select the highest-scoring result. Developers can fine-tune agent counts, voting thresholds, and model combinations to enhance reliability, mitigate individual bias, and refine solution quality. Swarm supports chaining responses, iterative feedback loops, and detailed reasoning logs for auditability, elevating performance on summarization, classification, code generation, and complex reasoning tasks through collective intelligence.

SuperSwarm is designed for orchestrating AI-driven workflows by leveraging multiple specialized agents that communicate and collaborate in real time. It supports dynamic task decomposition, where a primary controller agent breaks down complex goals into subtasks and assigns them to expert agents. Agents can share context, pass messages, and adapt their approach based on intermediate results. The platform offers a web-based dashboard, RESTful API, and CLI for deployment and monitoring. Developers can define custom roles, configure swarm topologies, and integrate external tools via plugins. SuperSwarm scales horizontally using container orchestration, ensuring robust performance under heavy workloads. Logs, metrics, and visualizations help optimize agent interactions, making it suitable for tasks like advanced research, customer support automation, code generation, and decision-making processes.

AgentSpeak is an open-source Java-based implementation of the AgentSpeak(L) programming language, designed to facilitate the creation and management of BDI (Belief-Desire-Intention) autonomous agents. It features a runtime environment that parses AgentSpeak(L) code, maintains agents’ belief bases, triggers events, and selects and executes plans based on current beliefs and goals. The interpreter supports concurrent agent execution, dynamic plan updates, and customizable semantics. With a modular architecture, programmers can extend core components such as plan selection and belief revision. AgentSpeak enables developers in academia and industry to prototype, simulate, and deploy intelligent agents in simulations, IoT systems, and multi-agent scenarios.

The Berkeley Pacman Projects repository offers a modular Python codebase where users build and test AI agents in a Pacman maze. It guides learners through uninformed and informed search (DFS, BFS, A*), adversarial multi-agent search (minimax, alpha-beta pruning), and reinforcement learning (Q-learning with feature extraction). Integrated graphical interfaces visualize agent behavior in real time, while built-in test cases and an autograder verify correctness. By iterating on algorithm implementations, users gain practical experience in state space exploration, heuristic design, adversarial reasoning, and reward-based learning within a unified game framework.

UnitMesh Framework provides a flexible, modular environment for defining, managing, and executing chains of AI agents. It allows seamless integration with OpenAI, Anthropic, and custom models, supports Python and Node.js SDKs, and offers built-in memory stores, tool connectors, and plugin architecture. Developers can orchestrate parallel or sequential agent workflows, track execution logs, and extend functionality via custom modules. Its event-driven design ensures high performance and scalability across cloud and on-premise deployments.

ElizaOS provides a robust suite of tools to design, test, and deploy autonomous AI agents within TypeScript projects. Developers define agent personalities, goals, and memory hierarchies, then leverage ElizaOS's planning system to outline task workflows. Its modular connector architecture simplifies integrating with communication platforms—Discord, Telegram, Slack, X—and blockchain networks via Web3 adapters. ElizaOS supports multiple LLM backends (OpenAI, Anthropic, Llama, Gemini), allowing seamless switching between models. Plugin support extends functionality with custom skills, logging, and observability features. Through its CLI and SDK, teams can iterate on agent configurations, monitor live performance, and scale deployments in cloud environments or on-premises. ElizaOS empowers companies to automate customer interactions, social media engagement, and business processes with autonomous digital workers.

EthicalEvalMAS provides a modular environment to assess multi-agent systems across key ethical dimensions such as justice, autonomy, privacy, transparency, and beneficence. Users can generate custom scenarios or use built-in templates, define bespoke metrics, execute automated evaluation scripts, and visualize outcomes through built-in reporting tools. Its extensible architecture supports integration with existing MAS platforms and facilitates reproducible ethical benchmarking across different agent behaviors.

The IntelligentMASPlatform is built to accelerate development and deployment of multi-agent systems by offering a modular architecture with distinct agent, environment, and service layers. Agents communicate using FIPA-compliant ACL messaging, enabling dynamic negotiation and coordination. The platform includes a versatile environment simulator allowing developers to model complex scenarios, schedule agent tasks, and visualize agent interactions in real-time through a built-in dashboard. For advanced behaviors, it integrates reinforcement learning modules and supports custom behavior plugins. Deployment tools allow packaging agents into standalone applications or distributed networks. Additionally, the platform's API facilitates integration with databases, IoT devices, or third-party AI services, making it suitable for research, industrial automation, and smart city use cases.

LinkAgent provides a lightweight microkernel for building AI agents with pluggable components. Users can register language model backends, retrieval modules, and external APIs as tools, then assemble them into workflows using built-in planners and routers. LinkAgent supports memory handlers for context persistence, dynamic tool invocation, and configurable decision logic for complex multi-step reasoning. With minimal code, teams can automate tasks like QA, data extraction, process orchestration, and report generation.

MACL is a modular Python framework designed to simplify the creation and orchestration of multiple AI agents. It lets you define individual agents with custom skills, set up communication channels, and schedule tasks across an agent network. Agents can exchange messages, negotiate responsibilities, and adapt dynamically based on shared data. With built-in support for popular LLMs and a plugin system for extensibility, MACL enables scalable and maintainable AI workflows across domains like customer service automation, data analysis pipelines, and simulation environments.

MultiAgent2 provides a comprehensive toolkit for orchestrating autonomous AI agents powered by large language models. Developers can define agents with customizable personas, strategies, and memory contexts, enabling them to converse, share information, and collectively solve problems. The framework supports pluggable storage options for long-term memory, role-based access to shared data, and configurable communication channels for synchronous or asynchronous dialogue. Its CLI and Python SDK facilitate rapid prototyping, testing, and deployment of multi-agent systems for use cases spanning research experiments, automated customer support, content generation pipelines, and decision support workflows. By abstracting inter-agent communication and memory management, MultiAgent2 accelerates the development of complex AI-driven applications.

NeuralABM is an open-source Python library that leverages PyTorch to integrate neural networks into agent-based modeling. Users can specify agent architectures as neural modules, define environment dynamics, and train agent behaviors using backpropagation across simulation steps. The framework supports custom reward signals, curriculum learning, and synchronous or asynchronous updates, enabling the study of emergent phenomena. With utilities for logging, visualization, and dataset export, researchers and developers can analyze agent performance, debug models, and iterate on simulation designs. NeuralABM simplifies combining reinforcement learning with ABM for applications in social science, economics, robotics, and AI-driven game NPC behaviors. It provides modular components for environment customization, supports multi-agent interactions, and offers hooks for integrating external datasets or APIs for real-world simulations. The open design fosters reproducibility and collaboration through clear experiment configuration and version control integration.