Ruflo: The Agent Orchestration Platform That's Exploding on GitHub
Published: 2026-05-04 · 10 min read · GitHub: ruvnet/ruflo · 1,840+ stars today
⚡ Trending #1 on GitHub today — Ruflo turns Claude Code into a multi-agent orchestration platform. Deploy 100+ specialized agents, coordinate them in swarms, persist memory across sessions, and even connect agents running on different machines. All through a single CLI plugin. This is the most significant Claude Code extension to date.
What Is Ruflo?
Ruflo (formerly Claude Flow) is an agent orchestration platform built natively for Claude Code. At its core, it answers a simple question: what if Claude Code wasn't limited to a single agent, but could coordinate a whole team of specialized agents?
With 1,840+ stars in a single day, Ruflo has become the fastest-growing Claude Code extension on GitHub. It brings production-grade orchestration to Claude Code through:
- 100+ specialized agents for coding, testing, security, docs, architecture, and more
- Swarm coordination — hierarchical, mesh, and adaptive topologies with consensus
- Self-learning memory — agents remember across sessions via vector database
- Zero-trust federation — agents on different machines collaborate securely
- 32 native plugins installed through Claude Code's
/pluginsystem - Web UI — multi-model chat with 210+ MCP tools at flo.ruv.io
Under the hood, Ruflo uses WASM kernels written in Rust to power its policy engine, embeddings, and proof system. The CLI is an NPM package that registers as both a Claude Code plugin and an MCP server.
How it works: User → Ruflo (CLI/MCP) → Router → Swarm → Agents → Memory → LLM Providers
All connected by a self-learning feedback loop — agents improve over time through trajectory learning.
Why Ruflo Matters in 2026
2026 has been the year of the multi-agent system. By May, we've seen:
- DeepClaude rerouting Claude Code to cheaper backends
- Claude Code adopting native MCP server support
- The debate about whether agentic coding is the future or a trap
- GitHub Copilot Codex expanding into agent mode
Ruflo is the first platform that treats Claude Code not as a single tool, but as a runtime for a fleet of agents. Instead of context-switching between tools (this for testing, that for security, something else for docs), you install one plugin and get a whole team.
Quick Start: Install Ruflo in 30 Seconds
Option 1: One-line install (recommended)
curl -fsSL https://cdn.jsdelivr.net/gh/ruvnet/ruflo@main/scripts/install.sh | bash
Option 2: Via npx
npx ruflo@latest init --wizard
Option 3: Global install via npm
npm install -g ruflo@latest
Option 4: Add as MCP server in Claude Code
claude mcp add ruflo -- npx -y @claude-flow/cli@latest
Plugin Marketplace: 32 Ways to Extend Claude Code
Ruflo's plugin marketplace is its killer feature. After installation, plugins are added through Claude Code's native /plugin system:
/plugin marketplace add ruvnet/ruflo
/plugin install ruflo-core@ruflo
/plugin install ruflo-swarm@ruflo
/plugin install ruflo-autopilot@ruflo
Here's what the major plugin groups do:
Foundation & Coordination
| Plugin | What It Does |
|---|---|
ruflo-core |
Foundation — server, health checks, plugin discovery |
ruflo-swarm |
Coordinate multiple agents as a team |
ruflo-autopilot |
Let agents run autonomously in a loop |
ruflo-federation |
Agents on different machines collaborate securely |
ruflo-workflows |
Reusable multi-step task templates |
Memory & Intelligence
| Plugin | What It Does |
|---|---|
ruflo-agentdb |
Fast vector database (150x–12,500x faster search) |
ruflo-rag-memory |
Hybrid search, graph hops, diversity ranking |
ruflo-rvf |
Save and restore agent memory across sessions |
ruflo-knowledge-graph |
Build and traverse entity relationship maps |
ruflo-intelligence |
SONA neural patterns + trajectory learning |
Dev Tools & Quality
| Plugin | What It Does |
|---|---|
ruflo-testgen |
Find missing tests and generate them automatically |
ruflo-browser |
Automate browser testing with Playwright |
ruflo-jujutsu |
Analyze git diffs, score risk, suggest reviewers |
ruflo-docs |
Generate and maintain documentation automatically |
ruflo-security-audit |
Scan for vulnerabilities and CVEs |
Architecture & Methods
| Plugin | What It Does |
|---|---|
ruflo-adr |
Track architecture decisions with a living record |
ruflo-ddd |
Scaffold domain-driven design — contexts, aggregates, events |
ruflo-sparc |
Guided 5-phase development methodology with quality gates |
ruflo-plugin-creator |
Scaffold, validate, and publish your own plugins |
Operations & Observability
| Plugin | What It Does |
|---|---|
ruflo-migrations |
Manage database schema changes safely |
ruflo-observability |
Structured logs, traces, and metrics |
ruflo-cost-tracker |
Track token usage, set budgets, get cost alerts |
ruflo-wasm |
Run sandboxed WebAssembly agents |
Total: 32 native Claude Code plugins + 21 NPM plugins covering everything from IoT device management to AI trading agents.
Swarm Coordination: How Multi-Agent Collaboration Works
The most architecturally interesting aspect of Ruflo is its swarm coordination system. Agents aren't just running independently — they organize into teams with specific coordination patterns:
- Hierarchical swarms — a manager agent delegates to worker agents, collects results, and synthesizes. Good for structured tasks like "audit codebase and fix all security issues" where each agent checks different vulnerability classes.
- Mesh swarms — all agents communicate peer-to-peer. Good for brainstorming and research tasks where insights need to cross-pollinate.
- Adaptive topologies — Ruflo dynamically reorganizes agent relationships based on the task at hand, learned from previous successful patterns.
Each swarm uses consensus mechanisms to resolve conflicts when agents disagree — a pattern borrowed from distributed systems that prevents agentic chaos.
Self-Learning: Agents That Get Smarter Over Time
Ruflo includes a self-learning loop powered by SONA (Self-Optimizing Neural Architecture) patterns and trajectory learning:
- Every agent interaction is logged to a ReasoningBank
- Successful patterns are identified and reinforced
- Future calls to the same or similar tasks skip exploration and go straight to the proven approach
- Memory persists across Claude Code sessions through the vector-backed AgentDB
This means the more you use Ruflo, the faster it gets. A task that might take 5 agent turns on day one could resolve in 1-2 turns after the system has learned the optimal approach.
Agent Federation: Cross-Machine Collaboration
The most futuristic feature is zero-trust agent federation. Agents running on different machines (your laptop, a CI server, a cloud VM) can discover each other, authenticate, and collaborate without leaking sensitive data.
This is built on Ruflo's WASM-based policy engine — each agent carries an encoded policy that defines what data it can share, with whom, and under what conditions. The routing layer enforces these policies automatically.
Use cases include:
- A local coding agent requesting a security audit from a locked-down CI security agent
- A development agent passing test results to a deployment agent running on a production server
- Teams of agents distributed across organizational boundaries collaborating on shared codebases
Ruflo Web UI: Multi-Model Chat with 210+ MCP Tools
Ruflo also includes a web-based interface at flo.ruv.io that provides:
- Multi-model chat — Qwen 3.6 Max, Claude Sonnet 4.6, Claude Haiku 4.5, Gemini 2.5 Pro, Gemini 2.5 Flash, OpenAI — available immediately via OpenRouter
- ~210 MCP tools — 5 server groups (Core, Intelligence, Agents, Memory, DevTools) plus an 18-tool browser-based gallery that works offline
- BYO MCP servers — click the "MCP" pill in chat input to add any MCP endpoint (HTTP, SSE, or stdio)
- Local LLM support — ruvLLM, Ruflo's self-improving local model layer that routes to MicroLoRA adapters
- No install, no API key required — just visit and try
Cost Analysis: Is It Worth It?
Ruflo itself is free and open-source (MIT license). The costs come from the LLM API calls your agents make:
| LLM Provider (via OpenRouter) | Cost per Token | Best For |
|---|---|---|
| Qwen 3.6 Max | Free / minimal | Routine coding |
| DeepSeek V4 Pro | $0.87/M output | Balance of cost & quality |
| Claude Sonnet 4.6 | $3.00/M output | Complex reasoning |
| Claude Opus 4.7 | $15.00/M output | Critical architecture work |
For heavy daily use with DeepSeek V4 Pro or Qwen, expect $20–60/month in API costs — significantly less than a dedicated Claude Code Pro subscription.
Limitations and Caveats
Ruflo is impressive, but it's worth being honest about its current state:
- Very new. The project rebranded from "Claude Flow" to "Ruflo" recently. APIs and plugin interfaces are likely to change.
- 32 plugins is a lot to manage. You probably don't need all of them. Start with core + swarm + one or two utility plugins.
- Dependency on OpenRouter. The web UI relies on OpenRouter for model access. If OpenRouter has an outage, the web UI is affected.
- Learning curve. While Ruflo promises "just use Claude Code normally," understanding which plugins to install and how to configure them takes time.
- Overhead. Running 100+ agent definitions locally adds startup time and memory overhead to Claude Code.
For most developers, the sweet spot is: install ruflo-core + ruflo-swarm + ruflo-rag-memory, add one or two dev tool plugins as needed, and skip the rest until you have a specific use case.
Conclusion
Ruflo represents a genuine leap forward for Claude Code. It's not just a plugin — it's a runtime for multi-agent systems that extends Claude Code from a single-agent tool into a full development platform. The 1,840+ stars in a day reflect genuine excitement, not hype.
Whether you're a solo developer who wants automated security scanning in your CI pipeline, or a team looking to deploy specialized agents across multiple machines with zero-trust federation, Ruflo has a plugin for that.
The era of single-agent coding tools is ending. Ruflo shows us what comes next.
Related Articles
- DeepClaude: Run Claude Code's Agent Loop with DeepSeek V4 Pro — 17x Cheaper
- Uncle Bob Says "It's Over" — AI Vibe Coding vs Traditional Software Engineering
- Claude Opus 4.7 发布全解读:xhigh 推理、代码工程与安全新特性
- MCP Server Config Setup Guide — Model Context Protocol for Claude Desktop & AI Agents
Published on EasyTool.me — developer tools and engineering guides. Last updated: 2026-05-04.