affaan-m/ECC
ECC is a harness-native operator system designed for agentic AI work, providing a comprehensive solution with skills, memory optimization, security, and research-first development across multiple A...
Awesome AI Agents › Multi-Agent Frameworks
Human-Agent-Society/CORAL
CORAL is open source infrastructure for running organisations of autonomous coding agents that carry out experiments, share knowledge and iteratively improve a solution, an approach its authors call autoresearch. The operator supplies a codebase and a grader, and the framework handles isolated workspaces, safe evaluation, persistent shared state and collaboration between agents. Agents run in parallel git worktrees and exchange findings through a common state directory holding attempts, notes and skills. Supported runtimes are Claude Code as the default, plus Codex, Cursor Agent, Kiro and OpenCode, each installed and authenticated separately and configurable per runtime, including through a LiteLLM gateway. Installation is a shell script that places the coral tool globally through uv, after which coral init scaffolds a task and coral start launches agents from a task.yaml file. Multi-island runs partition agents into isolated islands with scoped attempts, notes, skills and heartbeat state, and allow migration between islands to widen exploration. Grading is wired explicitly through a packaged grader entrypoint after the older auto-discovery mechanism was removed, and two reusable LLM-judge grader packages handle open-ended outputs such as reports, memos and legal analysis against a rubric. A Docker session mode isolates agents from the grader by running them as an unprivileged user while manager and grader stay root, so answer keys and the grader environment stay unreadable; on the host this isolation is opt-in. A plugin distributed through the repository marketplace lets an operator drive CORAL from inside Claude Code or Codex without learning the CLI: it is skills-first with no MCP, scaffolds a gitignored workspace, seeds the code, writes a grader and loops validation until the task is ready to launch, and on Claude Code a dedicated subagent can do that work autonomously while another triages stuck runs. The project is Apache 2.0, requires Python 3.11 or newer, and has an accompanying arXiv paper accepted at COLM 2026.
https://github.com/Human-Agent-Society/CORAL
Also in Multi-Agent Frameworks
ruvnet/ruflo
Ruflo is a leading agent orchestration platform for Claude, enabling the deployment of intelligent multi-agent swarms, coordination of autonomous workflows, and building conversational AI systems.
Yeachan-Heo/oh-my-claude-sisyphus
Oh My Claude Sisyphus (OMC) is a multi-agent orchestration framework for Claude Code, simplifying complex AI agent workflows without a steep learning curve.
microsoft/JARVIS
JARVIS is a collaborative AI system that connects large language models with expert AI models from HuggingFace to autonomously plan, execute, and integrate complex AI tasks.
rowboatlabs/rowboat
Rowboat is an AI-powered platform that enables users to build and manage multi-agent swarms using natural language, integrating tools and knowledge for automated workflows and deployment via API or SDK.
cft0808/edict
Edict is an OpenClaw Multi-Agent Orchestration System featuring 9 specialized AI agents, a real-time dashboard, model configuration, and full audit trails, inspired by ancient Chinese imperial gove...
MemoriLabs/Memori
Memori is a versatile SQL native memory layer designed to enhance LLMs, AI agents, and multi-agent systems with advanced memory management, seamless integration, and high performance.
HKUDS/DeepCode
DeepCode is an open-source project that uses multi-agent AI systems to convert research papers and natural language into production-ready code through both CLI and web interfaces.