// the find
ruvnet/ruflo
🌊 The leading agent orchestration platform for Claude. Deploy intelligent multi-agent swarms, coordinate autonomous workflows, and build conversational AI systems. Features enterprise-grade architecture, self-learning swarm intelligence, RAG integration, and native Claude Code / Codex Integration
Ruflo (formerly claude-flow) is a multi-agent orchestration layer built on top of Claude Code, adding swarm coordination, persistent vector memory, federated agent communication, and a plugin marketplace. It targets developers who want to run complex autonomous workflows where multiple specialized AI agents collaborate on tasks. The star count (45k+) is suspiciously high for a project of this maturity and scope.
- The two-path install model (lightweight Claude Code plugin vs full CLI with MCP server) is well-thought-out and honestly documented — the README explicitly tells you what each mode can't do, which is rare in this space.
- The federation architecture description (mTLS + ed25519, PII-gated data flow with 14-type detection, behavioral trust scoring) shows genuine thought about real security problems in multi-agent systems, not just checkbox security theater.
- Plugin system is modular and covers genuinely useful surface area — testgen, cost-tracker, observability, migration management — rather than just shipping one monolithic agent blob.
- HNSW-indexed AgentDB with sub-ms retrieval claims and explicit benchmarks ('150x-12,500x faster than brute force') is the kind of concrete performance framing that at least gives you something to verify.
- The star count (45k) vs forks (5k) ratio and the rename from claude-flow to ruflo mid-life raises questions about organic adoption vs. gaming; the README links to a different GitHub repo (ruvnet/claude-flow) in multiple badges while the repo is ruvnet/ruflo — that's a disorienting sign of hasty rebranding.
- Enormous surface area with very little verifiable substance: claims like '89% task routing accuracy', 'SONA neural patterns', 'self-learning', and 'Byzantine consensus' appear throughout but there are no benchmarks, test results, or reproducible experiments linked anywhere — the STATUS.md is vague and the verification.md is cryptographic hash checking, not capability verification.
- The federation feature is described in the README as 'See issue #1669 for complete architecture and implementation roadmap' — meaning a flagship capability exists primarily as a GitHub issue, not shipped code, which is a significant gap for something marketed as enterprise-grade.
- Extreme dependency on Anthropic's Claude Code plugin system means you're building on a surface that Anthropic can change or deprecate at will, and the 314 MCP tools / 98 agents scope means the maintenance burden for keeping all this working through API changes is enormous and almost certainly not being met.