swarmagent.dev
GitHub
DocsModelsResourcesPricingNewsContactGitHub

Open source · runs headless on your own machines

Your AI's team,built in.

Swarm runs headless on your own machines and gives your AI a coder, a finder, a designer and a reviewer. Tell your AI the workflow you want, like fixing failed builds, weekly posts or a morning brief, and Swarm runs it. You watch the whole fleet from the top.

Read swarmagent.dev/llms.txt and set up Swarm on this machine for me.

Learn MoreConnect your AIOpen source on GitHub

01 · Built in

The team is there the moment you install.

Swarm ships with its roles ready. Your AI hands Swarm a goal and it plans it and splits it across its coder, finder and designer, or gives one job straight to one of them. A reviewer checks work, and media makes images, video and sound. Your AI gets the result back.

  • Swarm, coder, finder, designer, reviewer, media: ready on install
  • Every job on its own branch: nothing lands until it's merged
  • Branch, diff, report and cost back to your AI

Built in · Media

Images, video and sound. Same box.

Ask for a product shot, a 10-second teaser or a jingle. Swarm sends it to an image, video or sound model you connected and saves the files into your project.

  • Create, edit or extend, starting from an earlier file
  • Checked against the model before anything is spent
  • Works with your ChatGPT plan, Google or OpenRouter

02 · Your workers

Your own workers, built on the roles.

When you'd otherwise explain the same thing every time, make a worker: a name, one of the roles and standing instructions it follows on every job. A site checker that is a finder, a release-notes writer that is a coder. Give it jobs yourself, or let a workflow run it.

  • A name, a role and standing instructions
  • Coder, designer, finder or reviewer underneath
  • Give it jobs, or put it on a schedule or trigger

03 · Workflows

Your workflow, running on its own.

Run a worker on cron, every few minutes, or whenever your app or CI calls it. Then decide what happens when any job ends: a reviewer checks it, it goes back for fixes, it's merged, or what it made is saved where you want it. Every run is kept.

  • Cron, interval or webhook triggers
  • After a job: review, send back, merge or deliver
  • Input, result and cost for every run

04 · Alerts

Know the moment something needs you.

Swarm writes a signal when a run fails, when a worker waits on you and when an agent is refused a connection. Send them to your own endpoint and turn them into alerts.

  • Signed deliveries (HMAC-SHA256)
  • A heartbeat, so a quiet box isn't mistaken for a dead one
  • One call answers: is the box OK?

05 · Chat bots and custom agents

A chat bot on your site. A person can step in.

Put a support desk, an intake form or an assistant on your site or in your app. It is a sealed Swarm agent: no shell, no files, no web, only the tools your app defines. Your server holds the one token that reaches it and decides who each visitor is, and someone from your team can take over a chat and hand it back.

  • Sealed: it can call only the tools your app answers
  • Your server decides who a visitor is, never the browser
  • A person takes over a chat and hands it back
  • Rate limits and a daily spending limit per bot; TypeScript SDK: @swarm-agent/sdk

06 · Machines

Headless on every machine. Watched from the top.

Swarm runs headless: a service on each machine, with no app to keep open and no screen. Put it on your PC, a VPS and a VM in your cloud account. Your AI sees all of them through one connection, gives each job to the right machine and watches the whole fleet from the top.

  • Tailscale: your machines find each other privately, no keys and no public ports
  • One MCP for every machine: each tool takes the machine to run on
  • Fleet monitoring: status, running jobs, today's spend and limits for every machine in one call
  • One model table for the whole fleet, so every machine works the same way
  • Results from every machine in one GCS or S3 bucket; claude.ai and ChatGPT on the web through your own relay preview
Hosted Swarm · coming soonWe run it for youNo server of your own needed. Leave your email and we'll tell you when it's ready →Open source · freeRun it yourselfInstall Swarm in your own cloud account or VPS and pay nothing for Swarm itself →

Your workflows

Build any workflow. Just describe it.

Your AI creates the workflow, connects the workers and puts it to work. Change it whenever you want.

competitor-watchhome-pc
  1. Every day08:00 · your timezone
  2. Finderresearches five competitors
  3. Reportwhat changed, sent to you

A research worker with standing instructions, on a daily schedule.

  • swarm_deploy_worker
  • swarm_schedule_job
  1. You invent the workflow.In a sentence, and change it with another.
  2. Your AI builds and manages it.Workers, instructions, schedules, triggers, machines.
  3. Swarm runs it.Every run on its own branch, kept and watched.

FAQ

Common questions.

What is Swarm?

Swarm runs AI agents on your own Linux machine, and your AI (Claude, ChatGPT or Codex) tells them what to do. It delegates a goal to Swarm, which splits it across its coder, designer and finder, or gives a job straight to one worker: code, apps, designs, research, images and video. It can put jobs on a schedule, add automations that run when a job ends, and give a folder its own environment. Then it reads what was delivered and what it cost, and merges the result. Swarm is open source under Apache 2.0.

Do I install it myself?

No. Paste one line into your AI (Claude, ChatGPT or Codex) on the machine. It installs Swarm, connects your model plan, sets up your projects and runs everything from then on. You talk to it in plain English.

What does it run on?

Ubuntu 24.04 on x86-64: your own PC, a server, a VPS or a cloud VM. Swarm doesn’t run on a Mac yet; from a Mac, your AI sets it up on a server and drives it from there.

Does it work on a Mac?

Not natively yet. Swarm needs Linux on x86-64, and an Ubuntu VM on an Apple silicon Mac is ARM, so it won’t install there. Get an Ubuntu 24.04 x86-64 server (a VPS, a cloud VM or a Linux PC at home) and tell your AI on the Mac: “Read swarmagent.dev/ai/mac.md and set up Swarm on a server for me.” It installs Swarm there over SSH and you keep working from the Mac.

Do I need an API key?

No. Your ChatGPT plan or your Claude plan (through Claude Code) works. You can also add API keys for OpenAI, Anthropic, Google, OpenRouter or Fireworks. You add keys yourself on the machine; they never go through chat.

Is it safe?

Swarm runs as its own user, listens on 127.0.0.1 only and opens no public ports. Each job works on its own branch, and nothing reaches your folder until your AI merges it. Secrets reach agents only through Swarm’s gateway, and agents only see a stand-in; keys never go through chat. Agents run their commands on the machine as Swarm’s user; for more isolation, turn on the gVisor sandbox or give a folder an environment (a container), and set spending limits.

Can I use it from my phone or laptop?

Yes, over Tailscale. Your AI on your other devices reaches Swarm over your private tailnet, with no public ports.

Can I build my own app on it?

Yes. Everything Swarm does is an MCP tool you can call over HTTP, so a program can do anything your AI can: give jobs, wait for them, read results and costs. Build your own app or orchestrator on top.

Can I run it in the cloud?

Yes. Install Swarm on any Ubuntu 24.04 VM or VPS in your own cloud account. Your AI reaches several machines as one fleet over Tailscale, or from claude.ai and ChatGPT on the web through your own Cloudflare relay (in preview); machines dial out, so nothing is exposed. Each machine can also export its activity and job results to one Google Cloud Storage or S3 bucket.

Can Swarm host this for me?

Hosted Swarm is coming soon, so you will not need your own server. Leave your email on the pricing page (swarmagent.dev/pricing) and we will tell you when it is ready. Until then, Swarm itself is free and open source to run on your own machines.

Give your AI a team.

Paste one line into your AI. It installs Swarm, runs it and tells you what happened.

Your first prompt

Read swarmagent.dev/llms.txt and set up Swarm on this machine for me.