Ruflo (Claude Flow) 🤖 Agent Free

What it does

Ruflo, the project formerly called Claude Flow, describes itself as an agent meta-harness for Claude Code and OpenAI Codex. It wraps the coding agent in an orchestration layer that spawns teams of specialist agents, coordinates them as swarms, stores what they learn in a vector memory and keeps that memory across sessions. The idea is that after setup you keep using Claude Code normally while hooks route tasks and coordinate agents in the background.

What is inside / how it works

The full install adds close to 100 agent definitions, dozens of commands and skills, an MCP server with several hundred tools, hooks and a background daemon. Swarms can run in hierarchical or mesh topologies with a coordinating "queen" agent and consensus protocols. Memory is handled by AgentDB with HNSW vector search, and a learning loop records successful patterns for reuse. Around 35 optional plugins add features such as test generation, security audits, architecture decision records, cost tracking, browser testing with Playwright, local model routing through Ollama and agent federation across machines. There is also a beta web chat and a hosted goal planner.

Works with

Claude Code is the primary host, and a separate Codex plugin package is published. The CLI can route to several model providers, including Claude, GPT, Gemini, Cohere and Ollama.

How to install or start

There are two paths. The lighter one adds individual plugins inside Claude Code with /plugin marketplace add ruvnet/ruflo. The full setup runs the init wizard, and Ruflo can also be registered as an MCP server:

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

The CLI install writes files such as .claude/, .claude-flow/ and CLAUDE.md into your project.

Pricing

Free and open source under the MIT licence. Swarms of parallel agents multiply model calls, so expect noticeably higher token bills than a single agent.

Who should use it

Experimenters and teams who want to push parallel agents, shared memory and automated workflows further than stock Claude Code allows. The surface area is very large and changes quickly, so it takes time to learn, and many performance claims in the README are the project's own. Start with a single plugin before committing to the full install.