RuFlo: Run 60+ AI Agents in Claude Code (No Coding Needed)

What RuFlo is

RuFlo is a free, open-source project that turns one Claude Code session into a swarm of sixty-plus agents working at the same time. It was built by ruvnet, it is MIT licensed, and it has over 65,000 stars on GitHub. There is nothing to buy.

An agent is a helper that goes off and finishes one job by itself. Picture handing a chore to your little brother — you are busy, he goes and does it, he comes back when it is done. The word you will see for that helper is an agent. Sixty of them means sixty jobs moving at once instead of you stepping through them one at a time.

Queen agents run the teams. Worker agents under them do the actual work. They all share one memory, so the swarm reuses whatever worked last time.

Plenty of people already run RuFlo. Almost none of them are people who don't write code, and that is backwards, because the jobs it is best at are not coding jobs.

What it actually saves you

Two things, and it is worth being precise about them because the numbers get thrown around loosely.

The first is routing. RuFlo reads how heavy each job is and sends it to the right model on its own, so light work goes to a cheap model and heavy work goes to a powerful one. You do not set this up and you do not think about it. RuFlo's docs say this "can extend your Claude Code usage by 250%."

The second is tokens. A token is a chunk of text that Claude reads or writes, and your plan is metered in tokens, so fewer tokens means more work before you hit a limit. RuFlo's docs put the savings there at 30 to 50 percent.

Both of those are RuFlo's own numbers, not measurements I have run. What you actually see moves with the kind of work you give it.

Before you start

• A terminal — texting your computer instead of pointing and clicking at it. On a Mac it is called Terminal, on Windows it is PowerShell. Both come installed already, so there is nothing to download.
• Node.js — the engine everything else runs on. Get the LTS version, which is the stable one.
• Claude Code, already working — RuFlo sits on top of Claude Code, it does not replace it.
• A paid Claude plan — RuFlo itself is free. Claude Code is the part that costs money.

If Claude Code is not set up yet, that is these two lines:

npm install -g @anthropic-ai/claude-code
claude

Which install to pick

RuFlo has two install paths and they are not the same size. One of them quietly removes the thing you came here for.

RuFlo's own README says the plugin path "adds slash commands and agent definitions only" and that its server is not registered. The hooks are the piece that reads each job and picks the model, which is where both numbers above come from. Skip the plugin and use the wizard.

Make a folder for this first. The wizard writes real files where it lands — a .claude folder, a .claude-flow folder, and a CLAUDE.md — so give it a home of its own instead of running it in Documents where the rest of your life lives.

mkdir -p ~/Documents/claude-projects/first-swarm
cd ~/Documents/claude-projects/first-swarm

Then run the wizard. It asks a few questions and sets everything up:

npx ruflo@latest init wizard

You run that once per folder, not once per machine.

Open Claude Code and talk to it the way you always do. RuFlo sits underneath and splits the work across the agents on its own. You never call them by name.

The one thing to know: give it a job big enough to be worth splitting. Here is one that runs on any machine with nothing to set up:

Go through my Downloads folder and sort every file into folders by what it
actually is - invoices, contracts, screenshots, receipts. Have one agent read
and sort, one rename them so I can find them later, and one check that nothing
got dropped. Show me the plan before you move anything.

That last sentence matters. Ask for the plan first on anything that touches your files.

Here is a bigger one:

Take the transcript in transcript.txt and turn it into a week of content -
five LinkedIn posts, ten X posts, and three email subject lines. Run the
writing agents at the same time, keep my voice from voice.md, and have a
reviewer agent kill anything that sounds like AI wrote it.

Both of those are three different roles working at once. That is the shape you are looking for.

On every run the swarm writes down what worked and reads it back on the next one. So the second thing you ask for in a folder tends to come out better than the first.

That is the argument for staying in one folder instead of starting fresh every time. The memory is the part that compounds.

Three things that trip people up

• Installing the plugin instead of running the wizard — the plugin skips the hooks, and the hooks are what do the routing. You end up with slash commands and none of the savings.
• Running it in a folder full of your stuff — the wizard writes several files and folders wherever you point it. Use an empty folder the first time so you can see exactly what showed up.
• Asking for something too small — a one-line fix has nothing to split across sixty agents. It is the big, boring, many-part jobs where the swarm actually pulls ahead.

Frequently asked questions

Do I need to know how to code to use RuFlo?

No. There is one setup command to run, and after that you type what you want in plain English. The two examples above are file cleanup and content writing, not programming.

Is RuFlo free?

Yes. It is open-source under the MIT license, with nothing to subscribe to. Claude Code itself still needs a paid Claude plan.

Should I use the plugin or the wizard?

The wizard. The Claude Code plugin adds slash commands but does not install the hooks that route each job to the right model, which is where the savings come from.

How much does RuFlo actually save?

RuFlo's docs put the token savings at 30 to 50 percent and say the routing can extend your Claude Code usage by 250%. Those are their numbers, and what you see depends on the work you give it.

Does it replace Claude Code?

No. It sits on top of Claude Code, so you keep working the way you already do.

Get RuFlo

The repo, the full docs, and the whole agent list live here: github.com/ruvnet/ruflo (opens in a new tab)