LogoAIAny
Icon for item

Claude-Flow

Claude-Flow is an enterprise-ready multi-agent orchestration framework (by ruvnet) that turns Claude Code into a coordinated multi-agent platform. It supports 54+ specialized agents, swarm topologies, RAG and vector memory (HNSW/AgentDB), self-learning components (SONA, EWC++), multi-LLM routing, MCP server integration and production security features — intended for complex software engineering, automation and AI-driven workflows.

Visit Website
Visit Website

Introduction

Claude-Flow — Enterprise AI orchestration for Claude

Claude-Flow (ruvnet) is a production-focused agent orchestration framework that extends Claude Code into a multi-agent development and execution platform. It provides a full stack for deploying coordinated agent swarms, persistent vector memory, model/provider routing, and learning-backed hooks so teams can automate complex engineering workflows reliably in production.

Core capabilities
  • 54+ specialized agents (coder, tester, reviewer, security architect, etc.) ready for common development tasks.
  • Swarm coordination with multiple topologies (hierarchical, mesh, ring, star, hybrid) and consensus support for fault tolerance.
  • Persistent memory and retrieval using HNSW/AgentDB for fast semantic lookups and pattern storage.
  • Self-learning intelligence: SONA + EWC++ pipeline for pattern distillation, preventing catastrophic forgetting and improving routing over time.
  • Multi-provider LLM support with intelligent routing and automatic failover (Claude, OpenAI, Google, local models).
  • Native MCP server integration for Claude Code (MCP tools, CLI and SDK) so the platform can be used from Claude, IDEs and other MCP clients.
  • Security features including AIDefence-style prompt/jailbreak detection, input validation, sandboxing and CVE-aware practices.
  • Plugin system and WASM extensions (RuVector / Agent Booster) for high-performance transforms and custom integrations.
Typical usage & integration
  • Run interactive tooling via CLI and MCP (e.g. bootstrap with npx claude-flow@v3alpha init, start MCP server with npx claude-flow@v3alpha mcp start).
  • Use prebuilt skills for code review, test generation, refactors, CI/CD automation and security audits.
  • Persist learned patterns and share/import pattern packs via a decentralized pattern marketplace (IPFS-enabled).
Technical highlights
  • Architecture: User → Claude-Flow (CLI / MCP) → Router → Swarm → Agents → Memory → LLM Providers, with a learning loop (RETRIEVE → JUDGE → DISTILL → CONSOLIDATE).
  • Embeddings & memory: ONNX/local embedding options and HNSW indices for sub-ms retrieval; optional centralized RuVector Postgres for large swarms.
  • Optimization: Agent Booster (WASM) for mechanical edits (avoid LLM costs), model tiering (haiku/sonnet/opus) to reduce API spend.
Quick install (example)
# Install or run via npx
npx claude-flow@v3alpha init
npx claude-flow@v3alpha mcp start
# Spawn an agent to implement a task
npx claude-flow@v3alpha --agent coder --task "Implement user authentication"
Who maintains it

The project is published under the GitHub organization/user ruvnet (homepage/reference: ruv.io). The repository includes detailed docs, MCP tools, CLI commands and an extensive README describing architecture, hooks, skills and deployment patterns.

Note: this summary is based on the project's README and repository metadata (created at 2025-06-02). For the latest releases, compatibility notes or governance details consult the repository and the project website.

Back

Information

  • Websitegithub.com
  • Authorsruvnet
  • Published date2025/06/02

Categories

Tags

More Items

Icon for item

Tianshou

2018
Jiayi Weng, Huayu Chen +8

Tianshou is a high-performance, modular deep reinforcement learning library built on pure PyTorch. It offers both high-level APIs for easy application and detailed procedural APIs for algorithm development. Tianshou supports online/offline RL, multi-agent setups, many standard algorithms (DQN, PPO, SAC, TD3, CQL, etc.), vectorized environments, EnvPool integration, recurrent models, multi-GPU training, and logging integrations (TensorBoard, W&B). It emphasizes software quality and reproducible results.

RLai-librarygithubai-trainai-framework+1
Icon for item

UltraRAG

2025
OpenBMB, THUNLP +2

UltraRAG is a lightweight RAG (Retrieval-Augmented Generation) framework based on the Model Context Protocol (MCP). It provides low-code orchestration of complex RAG pipelines by treating core components (retriever, generator, etc.) as independent MCP Servers and using an MCP Client for workflow composition. UltraRAG includes a visual Pipeline Builder, unified evaluation workflows, and one-click conversion of pipelines into interactive web demos, targeting both research exploration and industrial prototyping.

mcpRAGmcp-servermcp-clientai-workflow+3
Icon for item

Anthropic Sandbox Runtime (srt)

2025
Anthropic, anthropic-experimental (GitHub)

Anthropic Sandbox Runtime (srt) is a lightweight OS-level sandboxing tool that enforces filesystem and network restrictions on arbitrary processes without requiring full containers. It uses native primitives (sandbox-exec on macOS, bubblewrap on Linux) and proxy-based network filtering to limit what processes — including AI agents or MCP servers — can read, write, or connect to.

anthropicsecuritymcp-servermcpai-agent+4