ruvnet/claude-flow
Claude-Flow is an enterprise-grade AI orchestration platform that enables deployment and coordination of intelligent multi-agent swarms and autonomous workflows for building advanced conversational AI systems.
Awesome AI Agents › AI Agent Orchestration
ruvnet/agentic-flow
Agentic-Flow is a TypeScript orchestration layer for AI coding agents, distributed on npm and driven mainly through its CLI. Setup is three commands: initialize the project, pretrain from the existing codebase, then start Claude Code with the installed hooks. After that the project carries a routing layer that dispatches a natural-language task to the most appropriate agent from a catalog the README puts at 66 to 80 specialized roles including coder, tester, reviewer and architect, returning the chosen agent together with a confidence score. The same functionality is available as a library through an AgenticFlow class for programmatic routing. Alongside routing, the package registers a large set of Model Context Protocol tools, documented as 213, for swarm coordination, and can run as an MCP server that Claude Code connects to. Background workers are triggered by keywords such as ultralearn, optimize and audit, letting long-running analysis proceed outside the interactive session, and a metrics command reports how routing quality changes over time. Version 2 adds SONA, a self-optimizing component intended to learn routing patterns with sub-millisecond overhead, using LoRA fine-tuning for a research profile and EWC-style regularization to limit catastrophic forgetting while continual learning proceeds. It is paired with AgentDB, a vector and graph store whose release notes describe sparse and fused attention kernels, graph partitioning for memory reduction, GNN-based query refinement and WASM plus NAPI bindings. A model router selects cheaper backends where possible, which connects to the stated aim of letting Claude Code and Claude Agent SDK users switch to lower-cost alternatives such as OpenRouter or Gemini models and then deploy the resulting agents as hosted services. Note that the performance figures above are the maintainer claims from the README rather than independent measurements.
https://github.com/ruvnet/agentic-flow
Also in AI Agent Orchestration
code-yeongyu/oh-my-openagent
Oh-My-OpenAgent (OmO) is an advanced AI agent harness designed for complex codebases, orchestrating various AI models like Codex and OpenCode for software engineering tasks.
herdrdev/herdr
Herdr is an agent multiplexer that provides a runtime environment for coding agents directly within your terminal, offering a comprehensive view of agent activities and persistent sessions.
Donchitos/Claude-Code-Game-Studios
Transforms Claude Code into a comprehensive game development studio powered by 49 AI agents, 73 workflow skills, and a hierarchical coordination system for structured game creation.
BloopAI/vibe-kanban
Vibe Kanban is a Kanban board tool designed to manage and orchestrate AI coding agents, enabling efficient task tracking and workflow optimization in AI-assisted software development.
slopus/happy
Happy is an open-source harness orchestrating Claude Code, Codex and Grok sessions from one mobile/desktop app, reusing existing subscriptions, with end-to-end encrypted, multiplayer sessions.
getpaseo/paseo
Paseo is a self-hosted platform that orchestrates multiple coding agents like Claude Code, Codex, and Copilot across desktop and mobile devices, providing a unified interface for parallel execution...
NVIDIA/NemoClaw
NVIDIA NemoClaw is an open-source reference stack enabling the secure and sandboxed operation of AI agents like OpenClaw and Hermes within NVIDIA OpenShell environments, offering managed inference ...