Ruflo
Agent meta-harness for Claude Code and Codex that deploys intelligent multi-agent swarms with adaptive memory, self-learning intelligence, RAG integration, and federated coordination across machines.
Overview
Ruflo is an agent meta-harness for Claude Code and Codex — the execution layer that wraps AI coding tools with swarm coordination, persistent vector memory, and cross-machine agent federation. Where Claude Code writes code, Ruflo gives it a nervous system: 100+ specialized agents self-organize into swarms, learn from every task via SONA neural patterns, share memory across sessions via HNSW-indexed AgentDB, and — with federation — securely collaborate with agents on other machines without leaking data. One npx ruflo init installs 27 hooks, a full MCP server, and background workers that automatically route tasks and optimize patterns.
The Verdict
Who Should Use Ruflo?
Best For
- Claude Code and Codex users who want multi-agent coordination
- Teams building production AI workflows needing persistent memory
- Projects requiring swarm intelligence across many specialized agents
- Enterprises needing cross-machine agent federation with zero-trust security
- TypeScript developers extending Claude Code with custom plugins
Not Ideal For
- Developers happy with standard Claude Code for simpler coding tasks
- Python-first teams who don't use Claude Code or Codex
- Projects wanting a minimal, low-overhead setup
What's Great
- Turns Claude Code into a full multi-agent system with one install command
- HNSW vector memory measured 1.9x–4.7x faster than brute-force at scale (recall@10 ~0.99)
- 35 native Claude Code plugins covering swarms, RAG, security, observability, and trading
- Zero-trust federation lets agents on different machines collaborate without sharing raw data
- Self-learning SONA architecture improves task routing over time (89% routing accuracy claimed)
- MIT licensed, fully open-source with active development
Watch Out For
- Full CLI install adds significant workspace files (.claude/, CLAUDE.md, helpers)
- Steep learning curve — 60+ commands, 30 skills, and 314 MCP tools
- Plugin-only install doesn't register the MCP server, limiting many features
- Project was previously called Claude Flow — some docs still reference the old name
Pricing
View all features & details
Swarm & Coordination
- Hierarchical, mesh, and adaptive swarm topologies
- Queen-led consensus (Raft, Byzantine, Gossip protocols)
- 12 auto-triggered background workers
- Intelligent task routing (89% accuracy per benchmarks)
- Goal-Oriented Action Planning (GOAP) at goal.ruv.io
Memory & Intelligence
- HNSW-indexed AgentDB vector memory
- SONA self-learning neural patterns
- ReasoningBank trajectory learning
- Hybrid search with graph hops and diversity ranking
- Persistent memory across sessions
Multi-Provider LLM Support
- Claude, GPT, Gemini, Cohere, Ollama with smart routing
- Automatic failover between providers
- Local LLM support via ruvLLM and Ollama
- MicroLoRA adapter routing
Security & Compliance
- AIDefence prompt injection blocking
- PII detection and redaction (14-type pipeline)
- CVE scanning and remediation
- Zero-trust federation with mTLS + ed25519
- HIPAA, SOC2, GDPR audit trails
How It Compares
| Feature | Ruflo | Claude Code Alone | AutoGen | CrewAI |
|---|---|---|---|---|
| Agent Coordination | Swarm + federation | None | Multi-agent | Role-based |
| Persistent Memory | HNSW vector DB | Session only | Limited | Limited |
| LLM Providers | 5 with failover | Anthropic only | Multiple | Multiple |
| Self-Learning | SONA patterns | None | None | None |
| Claude Code Plugins | 35 native | External only | N/A | N/A |
| Cross-Machine | Zero-trust federation | None | Limited | None |