CommNet is a research-oriented library that implements the CommNet architecture, allowing multiple agents to share hidden states at each timestep and learn to coordinate actions in cooperative environments. It includes PyTorch model definitions, training and evaluation scripts, environment wrappers for OpenAI Gym, and utilities for customizing communication channels, agent counts, and network depths. Researchers and developers can use CommNet to prototype and benchmark inter-agent communication strategies on navigation, pursuit–evasion, and resource-collection tasks.

ePH-MAPF provides an efficient pipeline for computing collision-free paths for dozens to hundreds of agents on grid-based maps. It uses prioritized heuristics, incremental search techniques, and customizable cost metrics (Manhattan, Euclidean) to balance speed and solution quality. Users can select between different heuristic functions, integrate the library into Python-based robotics systems, and benchmark performance on standard MAPF scenarios. The codebase is modular and well-documented, enabling researchers and developers to extend it for dynamic obstacles or specialized environments.

MAGAIL implements a multi-agent extension of Generative Adversarial Imitation Learning, enabling groups of agents to learn coordinated behaviors from expert demonstrations. Built in Python with support for PyTorch (or TensorFlow variants), MAGAIL consists of policy (generator) and discriminator modules that are trained in an adversarial loop. Agents generate trajectories in environments like OpenAI Multi-Agent Particle Environment or PettingZoo, which the discriminator uses to evaluate authenticity against expert data. Through iterative updates, policy networks converge to expert-like strategies without explicit reward functions. MAGAIL’s modular design allows customization of network architectures, expert data ingestion, environment integration, and training hyperparameters. Additionally, built-in logging and TensorBoard visualization facilitate monitoring and analysis of multi-agent learning progress and performance benchmarks.

Patrolling-Zoo offers a flexible framework enabling users to create and experiment with multi-agent patrolling tasks in Python. The library includes a variety of grid-based and graph-based environments, each simulating surveillance, monitoring, and coverage scenarios. Users can configure the number of agents, map size, topology, reward functions, and observation spaces. Through compatibility with PettingZoo and Gym APIs, it supports seamless integration with popular reinforcement learning algorithms. This environment facilitates benchmarking and comparing MARL techniques under consistent settings. By providing standard scenarios and tools to customize new ones, Patrolling-Zoo accelerates research in autonomous robotics, security surveillance, search-and-rescue operations, and efficient area coverage using multi-agent coordination strategies.

A2A (Agent-to-Agent Architecture) is a Google open-source framework enabling the development and operation of distributed AI agents working together. It offers modular components to define agent roles, communication channels, and shared memory. Developers can integrate various LLM providers, customize agent behaviors, and orchestrate multi-step workflows. A2A includes built-in monitoring, error management, and replay capabilities to trace agent interactions. By providing a standardized protocol for agent discovery, message passing, and task allocation, A2A simplifies complex coordination patterns and enhances reliability when scaling agent-based applications across diverse environments.

AI-Agents provides a modular architecture for defining and running Python-based AI agents. Developers can configure agent behaviors, integrate external APIs or tools, and manage agent memory across sessions. It leverages popular LLMs, supports multi-agent collaboration, and enables plugin-based extensions for complex workflows like data analysis, automated support, and personalized assistants.

Agents with ADK is an open-source Python framework designed to streamline the creation of intelligent agents powered by large language models. It includes modular agent templates, built-in memory management, tool execution interfaces, and multi-agent coordination capabilities. Developers can quickly plug in custom functions or external APIs, configure planning and reasoning chains, and monitor agent interactions. The framework supports integration with popular LLM providers and provides logging, retry logic, and extensibility for production deployments.

CArtAgO (Common ARTifact Infrastructure for AGents Open environments) is a lightweight, extensible framework for implementing environment infrastructures in multi-agent systems. It introduces the concept of artifacts: first-class entities representing environment resources with defined operations, observable properties, and event interfaces. Developers define artifact types in Java, register them in environment classes, and expose operations and events for agent consumption. Agents interact with artifacts using standard actions (e.g., createArtifact, observe), receive asynchronous notifications of state changes, and coordinate through shared resources. CArtAgO integrates easily with agent platforms such as Jason, JaCaMo, JADE, and Spring Agent, enabling hybrid system development. The framework provides built-in support for artifact documentation, dynamic loading, and runtime monitoring, facilitating rapid prototyping of complex agent-based applications.

Cooperative Search Environment provides a flexible, gym-compatible multi-agent reinforcement learning environment tailored for cooperative search tasks in both discrete grid and continuous spaces. Agents operate under partial observability and can share information based on customizable communication topologies. The framework supports predefined scenarios like search-and-rescue, dynamic target tracking, and collaborative mapping, with APIs to define custom environments and reward structures. It integrates seamlessly with popular RL libraries such as Stable Baselines3 and Ray RLlib, includes logging utilities for performance analysis, and offers built-in visualization tools for real-time monitoring. Researchers can adjust grid sizes, agent counts, sensor ranges, and reward sharing mechanisms to evaluate coordination strategies and benchmark new algorithms effectively.

OpenAI Agents MCP Server provides a robust foundation for deploying and managing autonomous agents powered by OpenAI models. It exposes a flexible RESTful API to create, configure, and control agents, enabling developers to orchestrate multi-step tasks, coordinate interactions between agents, and maintain persistent memory across sessions. The framework supports plugin-like tool integrations, advanced conversation logging, and customizable planning strategies. By abstracting infrastructure concerns, MCP Server streamlines the development pipeline, facilitating rapid prototyping and scalable deployment of conversational assistants, workflow automations, and AI-driven digital workers in production environments.

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.