Mindket

Ruflo

Newversion 2 · updated
Star81.2k

Overview

ruvnet’s free, MIT-licensed layer that adds agent swarms, shared memory, a console and an MCP server to Claude Code and Codex. It was called Claude Flow. The setup notes cover Claude Code, Codex, Cursor, Windsurf, VS Code, JetBrains, Claude Desktop, ChatGPT and Grok Build.

Ruflo is an open-source project on GitHub in the Developer category, listed by Mindket. It is free. This is version 2, last updated on . Install it from GitHub as its README says. Get Mindket’s setup notes on your Mindket account, as a zip.

At a glance

Runs in
  • Claude Code: plugins from the maker’s marketplace, the npx command line, or a bare MCP entry
  • Codex: init --codex, which writes AGENTS.md and registers the MCP server, plus a dual mode that runs Claude Code and Codex together
  • Cursor, Windsurf, VS Code and JetBrains: one MCP entry in each tool’s own file or settings
  • Claude Desktop: an MCP entry in its config file
Start with
/plugin marketplace add ruvnet/ruflo
/plugin install ruflo-core@ruflo
/plugin install ruflo-console@ruflo

Platform

Claude CodeCoding agent
CodexCoding agent
CursorCode editor
WindsurfCode editor
VS CodeCode editor
Claude DesktopDesktop app

And 1 more in Where to use it

About

What it is

Ruflo is a free, MIT-licensed layer that sits around Claude Code or Codex. The README splits the job in two. The model writes, and Ruflo gives it tools, memory, loops, sandboxes and controls. It lists more than a hundred specialized agents, swarms that coordinate them, memory that carries across sessions, a console, and federation between machines. The README says Reuven Cohen (rUv) built it, first as Claude Flow.

  1. You give Claude Code or Codex a goal
  2. Ruflo’s router and hooks pick the agents for the task
  3. A swarm of specialized agents splits the work
  4. Memory keeps what worked for the next task
  5. The console shows the run, its cost and its results

Who it suits

People who already work in Claude Code or Codex, or another MCP client, and want several agents on one job with shared memory and a view of cost. Skip it for small edits. The maker’s own guide says a single agent is often better for a small, straightforward change.

Who uses it, and how
IndustryWho uses itHow they use itExample
Software teamsDevelopers building a featureRun the feature template, which hands a task down a pipeline of agentsA developer sends the template the task “Add user auth”, and the User Guide says an architect, a coder, a tester and a reviewer take it in turn.
Security and reviewEngineers who audit codeRun the security template on a folderAn engineer points it at src/auth/, and the guide lists a scanner, an analyzer and a fixer as its steps.
Agent operatorsPeople who run several agentsWatch missions, agents, models and cost in the consoleA person opens /ruflo in Claude Code, asks Claude to create a mission to add a dark mode toggle, and checks it in Missions.
Teams on separate machinesTwo teams that share findingsUse federation to send tasks between machinesThe README’s example has two teams sharing fraud signals, with personal data stripped from the task before it leaves.

Where to use it

Ruflo works in many tools, and the setup notes have a page for each. The plugins, the console and the mods are Claude Code integrations. Everywhere else Ruflo arrives as an MCP server, and the README says hook support depends on the client.

  • Claude Code: plugins from the maker’s marketplace, the npx command line, or a bare MCP entry
  • Codex: init --codex, which writes AGENTS.md and registers the MCP server, plus a dual mode that runs Claude Code and Codex together
  • Cursor, Windsurf, VS Code and JetBrains: one MCP entry in each tool’s own file or settings
  • Claude Desktop: an MCP entry in its config file
  • ChatGPT: a hosted plugin named RuFlo Federation, or your own remote server over HTTP
  • Grok Build: init --grok, which writes a project folder of Grok settings and a team bus
  • A plain terminal: the optional console, started with npx ruflo tui

How to set it up

Get it saves a zip of Mindket’s setup notes in your downloads folder, with a page for each tool from the maker’s README and docs. Pick one route. The README’s table says the plugins put no files in your workspace, and the command line writes a full setup into it. The console and mods still keep their own state under .claude-flow/.

What you need first
NeedDetail
Node.js20 or newer. The User Guide lists it as required, and the npm entry says the same
Package managernpm 9 or newer, pnpm or bun, from the User Guide’s prerequisites
An agentClaude Code 2.1.287 or later for the plugin route (core, console and mods). Cursor, Windsurf, VS Code 1.102+, JetBrains and Claude Desktop start a local MCP server. ChatGPT needs the hosted plugin or a remote server
A model keyANTHROPIC_API_KEY for Claude models, from the User Guide. OpenAI and Google keys are optional
A shellThe curl installer needs a POSIX shell. On native Windows use the npx line
  1. Install Node.js 20 or newer, then your agent.
  2. In Claude Code, add the marketplace and install the plugins, or run the setup wizard in your project folder.
  3. In any other client, add the MCP server entry from its page in the notes.
  4. Restart the client, then ask it to list the connected tools and run one read-only call.
/plugin marketplace add ruvnet/ruflo
/plugin install ruflo-core@ruflo
/plugin install ruflo-console@ruflo
/plugin install ruflo-mods@ruflo
/reload-plugins
/ruflo

Type those inside Claude Code. For Codex and other MCP clients, the README runs the setup wizard in your project terminal.

npx ruflo@latest init wizard

The MCP server alone is one line, and the User Guide’s Cursor, Windsurf and VS Code files hold the same entry in JSON.

npx -y ruflo@latest mcp start
{
  "mcpServers": {
    "ruflo": {
      "command": "npx",
      "args": ["ruflo@latest", "mcp", "start"]
    }
  }
}

How to use it

After init, the README says to use Claude Code as you always do. Its hooks route tasks and coordinate agents in the background. You can also start agents yourself. These three lines are the User Guide’s basic usage.

npx ruflo@latest agent spawn -t coder --name my-coder
npx ruflo@latest hive-mind spawn "Implement user authentication"
npx ruflo@latest agent list

In Claude Code, /ruflo opens the console. Its Claude control setting has four levels, read, write, manage and full, each with ask or auto approval. The README says the default is read plus ask.

Example

The README walks through one task, creating a mission with Claude. Install the console, open /ruflo and set Settings, then Claude control, to write plus ask. Then send the README’s own prompt.

Create a mission to add a dark mode toggle. Show me the mission and its status.

Claude opens Missions and sets the goal, and you confirm the create action in the console. Check that Missions holds the goal and Overview records the action. The README adds that creating a mission does not mean the feature has been built. Take back control pauses Claude’s console tools.

For newcomers and experts

If you are new. A coding agent is a model plus the tools around it, and Ruflo is the second part. A swarm is a group of agents that coordinate through a lead and shared memory. MCP is how a client such as Cursor calls Ruflo’s tools. The maker’s guide starts smaller than any of that, with one skill.

npx skills add ruvnet/ruflo --skill ruflo

The command is skills, plural. The guide says the skill gives an agent instructions for working with Ruflo, and does not start a server or grant access to your files. Its first prompt to the agent is this.

Use the RuFlo skill to explain what is installed and what is not. Do not change files or connect services yet. Propose the smallest setup for a read only review of this project.

For experts. The README names hierarchical, mesh and adaptive swarm topologies with consensus, and vector memory in an HNSW-indexed AgentDB. It says the learning loop depends on the configured memory and the quality of feedback. The speed comparisons are the maker’s own, in a gist the README links, and these notes did not rerun them.

Tool groups cut the number of MCP tools a client loads, which the User Guide says reduces latency and token use. Command line arguments beat environment variables, and those beat the config file.

export CLAUDE_FLOW_TOOL_GROUPS=implement,test,fix,memory
export CLAUDE_FLOW_TOOL_MODE=develop

The ruflo-core launcher reads RUFLO_MCP_SKIP_NPX=1 to refuse the unpinned npx fallback, and RUFLO_MCP_CLI_OVERRIDE to point at a built CLI. RuFlo Explained pins the release it checked in its examples, so a later version can change a command.

Good to know

The maker’s docs flag most of this themselves.

  • Names: Ruflo was Claude Flow. The README’s star and issue links still point to the claude-flow repository, and the MCP key, several package names and some commands still say claude-flow
  • Docs drift: RuFlo Explained says some documentation describes different release snapshots, so check the installed CLI help when commands differ. The tool count differs by page, so npx @claude-flow/cli@latest mcp tools gives the live list
  • Files in your project: the command line route writes .claude/, .claude-flow/, CLAUDE.md, helpers and settings. RuFlo Explained says to review the changes and to try it in a disposable project first
  • Permissions: mods run with your account’s permissions and are not sandboxed, so read their code first. In the console, network, spending and destructive actions need confirmation even in auto
  • Keys: the User Guide says never hardcode API keys in config files checked into version control
  • Cost: RuFlo Explained says more agents are not automatically cheaper. Parallel work can cut waiting and still raise token use
  • Remote servers: for ChatGPT over HTTP, RuFlo Explained says not to expose a local command server to the internet without authentication, tool restrictions and an owner
  • Proof: a connected server is not a working one. Run one read-only call and keep the result before you add more. In the release RuFlo Explained checked, doctor --fix printed suggested commands, and the guide says not to describe it as automatically repairing everything
  • Other repositories: the marketplace lists agentic-qe-fleet (maintained by Agentic QE), and agentdb and agentic-flow (maintained in ruvnet’s own agentdb and agentic-flow repositories). The README says to install each from one marketplace only
  • Cloud agents: the optional ruflo-agent feature can run an agent in Anthropic’s cloud. It needs ANTHROPIC_API_KEY and beta access, and its results come back marked unverified
  • Not tested here: the federation security claims, the learning loop and the speed comparisons are the maker’s statements. Mindket did not run them
  • Removal: the docs give no uninstall steps, so Mindket’s notes list what to remove by hand
  • License: MIT. The LICENSE file reads Copyright (c) 2024-2026 ruvnet. Mindket wrote the notes, and ruvnet did not write or endorse them

README and files

README

Ruflo: setup notes

This kit holds setup notes only. Mindket wrote them. Ruflo, its plugins and its command line install from ruvnet's repository: https://github.com/ruvnet/ruflo

These notes are not made or endorsed by ruvnet. The README says "Built by Reuven Cohen (rUv), originally as Claude Flow", and that Claude Flow is now Ruflo. Its first line calls Ruflo "An agent meta-harness for Claude Code and Codex." The README is at https://github.com/ruvnet/ruflo/blob/main/README.md

License: MIT. The LICENSE file (https://github.com/ruvnet/ruflo/blob/main/LICENSE) reads "Copyright (c) 2024-2026 ruvnet".

What it is, in the maker's words

The README says: "Agent = Model + Harness. The model writes; the harness gives it tools, memory, loops, sandboxes, and controls so it can actually work. Ruflo is the harness". It lists 100+ specialized agents, coordinated swarms, self-learning memory, federated comms across machines and security guardrails. It also says you do not need to learn every tool: after init, use Claude Code normally and the hooks route tasks and coordinate agents in the background. Source: README

The maker's own guide, RuFlo Explained (https://github.com/ruvnet/ruflo/blob/main/docs/ruflo-explained.md), puts it plainly: "You still choose the goal. Your AI model still does the reasoning." It also advises one coordinator, a few independent tasks, a fixed budget and a clear finish line, and says a single agent is often better for a small, straightforward change.

What installs

It depends on the route you pick, and the README says the routes differ a lot. Source: README, Quick Start

  • Claude Code plugins: slash commands, a few skills and agent definitions per plugin. The README's table says they add no files to your workspace and install no hooks, but the ruflo-core plugin README says its PreToolUse, PostToolUse, PreCompact and Stop hooks load when you install it, and the console and mods keep state under .claude-flow/. Among ruflo's own plugins, ruflo-core registers the MCP server. The external agentic-qe-fleet, agentdb and agentic-flow plugins ship their own.
  • The command line (npx ruflo init): the README lists 98 agents, 60+ commands, 30 skills, the MCP server, hooks and a daemon, with files in .claude/, .claude-flow/, CLAUDE.md, helpers and settings.
  • An MCP server only (npx ruflo@latest mcp start), for any client that can start a local stdio server.

What you need first

Main install commands

The README's two first routes. The first is for Claude Code, typed inside a session. The second runs the setup wizard in your project's terminal and works for Codex and other MCP clients.

/plugin marketplace add ruvnet/ruflo
/plugin install ruflo-core@ruflo
npx ruflo@latest init wizard

Source: README, Get started with Ruflo

Per-agent notes

  • setup/claude-code.md: plugins, the console and mods, the command line and the MCP server
  • setup/codex.md: init --codex, the MCP entry and dual mode with Claude Code
  • setup/cursor.md
  • setup/windsurf.md
  • setup/vs-code.md
  • setup/jetbrains.md
  • setup/claude-desktop.md
  • setup/chatgpt.md: the hosted plugin and your own remote server
  • setup/grok-build.md
  • setup/terminal-ui.md: ruflo tui, the console in a plain terminal
  • setup/tools-and-environment.md: environment variables, tool groups and key handling
  • setup/federation.md: joining agents on other machines

Update or remove

The User Guide gives these update lines:

npx ruflo@latest init upgrade
npx ruflo@latest init upgrade --add-missing

The first updates helpers and the statusline and preserves your data. The second also adds skills, agents and commands that are new in a later version, without overwriting your own changes. Source: https://github.com/ruvnet/ruflo/blob/main/docs/USERGUIDE.md (Upgrading)

The README and the User Guide give no uninstall steps. To take it out by hand, these notes (Mindket's, not ruvnet's) go by what the docs say init writes: the .claude/, .claude-flow/ and CLAUDE.md files, plus the helpers and settings the README lists for the command line route. Remove the MCP entry with your client's own command. For a global install, npm uninstall -g ruflo is npm's own command, not one from the maker.

Things the maker's own docs flag

  • RuFlo Explained says "Some documentation describes different release snapshots, so check the installed CLI help when commands differ." The User Guide's own top line names an alpha release as the latest, and its tool counts differ from the README's. Source: https://github.com/ruvnet/ruflo/blob/main/docs/ruflo-explained.md
  • Mods run with your account's permissions and are not sandboxed, so read their code before you install them. Source: README
  • The README says its console recording is a real Claude Haiku session that created a mission and opened pages. It says the workflow animation uses sample data and the theme demo replays renderer frames, and that neither is evidence of completed swarm work or memory recall.
  • A connected server is not proof it works. The RuFlo Explained guide says to run one read-only call and save the result before you add more.

Sources

Files
README.md  6 KB
setup/chatgpt.md  3 KB
setup/claude-code.md  7 KB
setup/claude-desktop.md  2 KB
setup/codex.md  4 KB
setup/cursor.md  2 KB
setup/federation.md  3 KB
setup/grok-build.md  4 KB
setup/jetbrains.md  1 KB
setup/terminal-ui.md  3 KB
setup/tools-and-environment.md  3 KB
setup/vs-code.md  1 KB
setup/windsurf.md  1 KB

13 files, 42 KB, downloaded as a zip.

What’s new

  • Version 2 · 2026-11-09

    Updated by Mindket to the current version of this kit.

  • Version 1 · 2026-11-03

Compare with similar

Listed GitHub projects in the same category that share the most topics with this one.

ProductStarsActivityLanguageLicenseLast change
Ruflo This page81,187——MIT—
DifyShares agentic-framework · agentic-workflow · agentic-ai160,579————
CopilotKitShares ai-assistant · agentic-ai · agents38,557ActiveTypeScriptMIT
UI UX Pro MaxShares ai-skills · claude · codex146,944ActivePythonMIT
ArchifyShares dsh-plugin · codex · ai-agents94,756ActiveJavaScriptApache-2.0
SimShares agentic-workflow · agents30,071——Apache-2.0—
Numbers as Mindket last read them, from 2026-11-07 to 2026-11-09. Activity is read from each project’s last change on GitHub: Active within 30 days, Quiet within 180, Stale after that.

Reviews

No ratings yet

Write a review

No reviews yet.

What people say

Posts and comments that name this project on Hacker News, Bluesky, X, Reddit, Facebook and YouTube, found by a weekly search and on Trendshift. Mindket doesn’t write or check them. Open one to read it where it was posted.

ms7892Ruflo: An agent meta-harness for Claude Code and Codex3 pointsGitHub Trending🎉 Celebrating! 🎉 (500+ new stars) 📦 ruvnet / ruflo ⭐ 76,913 (+564) 🗒 TypeScript 🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federatio...4 likes · 1 reply@PunkChainerTu agente de trading no necesita “más prompts”. Necesita stack: memoria, skills, routing y orquestación. En GitHub hay cientos de repos gratis que dan superpoderes a Claude (y a tu flota de agentes). En Punkchainer no montamos un chat monolítico: research, riesgo y ejecución viven separados, y…chunpaiyangIs any one using ruflo?1 point · 1 reply@hydromo.bsky.socialGitHub Trending weekly Top 10 (as of Oct 24) 1 security-audit-skill · 2 rea · 3 claude-plugins-official · 4 ruflo · 5 orca · 6 opencode · 7 cursor/plugins · 8 android/skills · 9 knowledge-work-plugins · 10 hyperframes Use cases and links are on the cards. https://github.com/trending?since=weekly1 reply@Ldataengineer$81@hydromo.bsky.social이번 주(10/24 기준) GitHub 트렌딩 TOP10 1 security-audit-skill · 2 rea · 3 claude-plugins-official · 4 ruflo · 5 orca · 6 opencode · 7 cursor/plugins · 8 android/skills · 9 knowledge-work-plugins · 10 hyperframes 활용 한 줄과 링크 10개는 카드에 정리했습니다. https://github.com/trending?since=weekly1 reply@MohsinFBAMore AI agents means more handoffs to manage. Ruflo adds coordination and memory around coding agents like Claude Code and Codex. Worth exploring: https://t.co/5h4g4aY2tz What breaks first in your agent workflows: context, handoffs or verification?GitTrending_BlueSky🔥 Ruflo: An Agent Meta-Harness for Intelligent Workflows 📦 ruvnet/ruflo Ruflo is gaining traction as a versatile meta-harness for deploying intelligent multi-agent syst... #github #opensource #TypeScript 👉 https://www.gittrending.com/article/ruflo-an-agent-meta-harness-for-intelligent-workflows1 like@FrankLeeLifeGeneric AI output is the disease. These 10 open source projects are the cure. GitHub trending added ~58K stars to them this week. Each kills one kind of AI slop: design, code, security, SEO. Which one are you trying first? Bookmark this ⬇️デイリーGitHubトレンド今日のGitHubトレンド ruvnet/ruflo Claude CodeやCodexなどのAIエージェント向けに、コアツールや管理コンソール、ランタイムモッドを提供するメタハーネスです。 MCPサーバーを介して多様なクライアントと連携し、コストやイベントの可視化、機能拡張を実現します。1 like@ZubairThinksAIIf you write skills for your team, treat them like code. Check every tool name and command against the real list, and again when your tools change. A wrong name won't show up as an error. It shows up as an agent that quietly does less.

Searched

More from Mindket

See all from Mindket

anaksor-ocr

Sokheng · Kit

Reads Khmer and English text from documents, receipts and photos of signs, using a detector for text lines and a recogniser for each line. Suits developers who…

New

Qwen2.5-7B-Instruct as GGUF files in eight quantization sizes, to chat with locally through llama.cpp. For people who want a capable 7.6B model on their own co…

New

ACE-Step 1.5 XL SFT

ACE-Step · Kit

Generates music from text with a 4B diffusion decoder, 50 steps and guidance control, aiming for higher audio quality than the base. For creators with a GPU of…

New

Similar projects

See all in Developer

Dify

langgenius · Kit

LangGenius’s open-source platform for building LLM apps: visual workflows, RAG, model management and agents with a sandbox. Its difyctl tool and skill let Clau…

New161k

Hermes Agent

NousResearch · Kit

Nous Research’s open source agent that you run in a terminal, in a desktop app or from a chat app. It keeps memory, writes skills from experience, runs schedul…

New261k

UI UX Pro Max

nextlevelbuilder · Skill

Generates tailored design systems, including color palettes, typography pairings, and layout patterns, based on specific project requirements and industry stan…

New147k