Ruflo (formerly Claude Flow)

What it does

• Two install paths with very different surface: a lite Claude Code plugin set (slash commands, zero workspace files) versus the full CLI loop — the README's table counts 98 agents, 60+ commands, 30 skills, an MCP server, hooks and a daemon
• Self-organizing agent swarms with hierarchical, mesh and adaptive topologies, plus self-learning memory (AgentDB vector memory, trajectory learning)
• Zero-trust agent federation: agents on different machines discover each other and exchange work over mTLS + ed25519, with a 14-type PII pipeline and HIPAA/SOC2/GDPR audit modes
• Visual console (npx ruflo tui, or /ruflo inside Claude Code) with cost tracking, swarm view and Claude control permission levels (read/write/manage/full)

When to use it

When a single Claude Code session stops being enough: multi-agent work that needs shared memory and coordination, background workers on timers, cross-machine agent collaboration under an explicit security model, or an audit of your agent setup (MetaHarness). Multi-provider routing (Claude, GPT, Gemini, Cohere, Ollama) covers failover beyond Anthropic.

Getting started

Full loop: npx ruflo@latest init (wizard: npx ruflo@latest init wizard), or npm install -g ruflo@latest. Lite path, inside Claude Code:

/plugin marketplace add ruvnet/ruflo
/plugin install ruflo-core@ruflo

MCP server:

claude mcp add claude-flow -- npx ruflo@latest mcp start

Things to know

The README says it outright: mods run with your account's permissions and are not sandboxed — review their code before installing, and the plugin path needs Claude Code v2.1.287+. The two paths are not interchangeable: the plugin track registers MCP tools under mcp__plugin_ruflo-core_* names, not the bare tool names the CLI scaffold uses. Surface area is genuinely large (hundreds of tools, dozens of plugins); the README's own advice is to learn none of it up front — after init, just use Claude Code normally.