ruvnet/claude-flow

Claude Flow (Ruflo v3.5) is a production-grade multi-agent orchestration platform built for Claude Code. It coordinates swarms of 60+ specialized AI agents across 12 categories — covering development, swarm coordination, GitHub management, testing, and domain-specific tasks — using mesh, hierarchical, ring, and star topologies with Byzantine fault-tolerant consensus. The platform routes tasks through a 3-tier system: a WASM-based Agent Booster handles simple code transforms in under 1ms at zero API cost, mid-tier models (Haiku/Sonnet) handle medium tasks, and Opus tackles complex architecture work. This reportedly extends Claude Max quotas by 2.5x and cuts API costs by up to 75%. Built in TypeScript with Rust WASM kernels, it ships 215+ MCP tools, a self-learning Q-Learning router with 42+ skills, and HNSW vector search delivering 150x-12,500x faster retrieval than naive search. Supports multi-provider LLM failover across Claude, GPT, Gemini, Mistral, and local models via Ollama. Install via npx, npm, or bun; integrates with VS Code, Cursor, Windsurf, JetBrains, and Claude Desktop. Note: the performance claims (352x speedup, 75% cost reduction) are self-reported benchmarks, not independently verified. Practical value depends heavily on use case complexity.