Each one can do the work of a person. Keeping up with all of them became your job.
claude
codex
opencode
Every agent is in a different window.
Claude Code in one window. Codex in another. OpenCode in a third. Two projects, five terminals, and you’re the only thing that knows what any of them are doing.
You didn’t mean to become the router. It just happened.
What if one place knew about all of them?
agent: that-one-cli
alternatives: none · switching cost: everything
Your whole workflow is a bet on one vendor.
One price change, one outage, one model that quietly got worse. Everything you’ve built around that one CLI is hostage to it.
You didn’t choose lock-in. It just built up.
What if switching agents was a dropdown, not a migration?
❯ approve file edit? [y/n]
agent: blocked · waiting on: you
Your agent is stuck waiting for you.
“Can I edit this file?” It asked at 12:10. You were at lunch. It’s 12:50, and the smartest coding tool you own has spent forty minutes doing nothing.
What if you could tap Approve from your phone?
You run the smartest agents ever built.
Running them shouldn’t be the hard part.
That’s the part we fixed.
Intelligence doesn't scale. Coordination does.
That's how every great team works. The wins never came from smarter people. They came from giving smart people better systems.
Your agents have the smarts. They've just never had a system.
Sowebuiltit.
Tasks
Hand off the work. It runs on schedule.
Relay
Your agents can reach you anywhere.
Mesh
Your agents find each other.
Console
See and steer every agent you run.
A Night with DorkOS
11:14 PM
You hand off three tasks. Then you close the laptop.
A test suite that needs expanding. A dependency upgrade across two services. A refactor you’ve been putting off.
You type one command. TASKS schedules all three.
You close the laptop.
11:15 PM
The first agent starts before you finish brushing your teeth.
It reads the coverage report, identifies the gaps, starts writing tests.
You are brushing your teeth.
2:47 AM
Something breaks. An agent fixes it. You never wake up.
The dependency upgrade TASKS scheduled runs into failing tests. The agent reads the error, traces the cause, opens a fix. Tests go green.
Your phone buzzes once. A Telegram message from RELAY: “Tests were failing. Fixed. Change #247 ready for your review.”
You do not see it until morning.
2:48 AM
Two agents work the same service. Neither one is flying blind.
MESH keeps a live map of which agents exist and where they’re working. So when a second agent picks up a task in the same service, it can see the first one is already there.
Discovery you didn’t have to wire up yourself.
7:00 AM
You open your laptop to a full progress report.
You open your laptop. CONSOLE shows the night at a glance: three changes ready for review, one fix already merged, the refactor at 80%, waiting on a design question it queued for you.
7:04 AM
You’ve been awake four minutes. The system you designed worked eight hours.
You approve two changes. You request a revision on the third. You queue two more tasks for the day.
The system you designed has been productive for eight hours. You’ve been awake for four minutes.
DorkOS is open source, MIT licensed, and runs wherever you put it: your laptop, a VPS, a Raspberry Pi. Every line of code is readable. Every decision is documented. Nothing phones home.
It coordinates your agents. The thinking is theirs. The system is yours.
We built it for ourselves. Now it’s yours.
DorkOS started as the system we needed. One person, five projects, ten agents in ten different windows. We built it so we could build more, faster.
It’s how we ship everything else: one person and a fleet of agents, 44 public releases in five months, every decision out in the open.
And yes, we named it after ourselves. Dork is what they call someone who cares too much about the plumbing. We answer to it. You never have to.
An agent is an AI coding tool, like Claude Code, Codex, or OpenCode, that can read, write, and run code on your machine. DorkOS doesn't replace your agents. It gives them what they're missing to work when you're not watching: a schedule, a way to reach you, a record of everything they did, and a way to find and message each other.
Claude Code is the agent: the tool that thinks and writes code. DorkOS is the system around it. Without DorkOS, your agent stops the moment you close the terminal. With it, your agents run on a schedule, message you when something breaks, message each other, and pick up right where they left off.
No. DorkOS runs entirely on your own computer. Your session data stays on your machine, in each AI tool's own local files. There are no accounts, no cloud dependency, and nothing phoning home to check in.
Node.js 22 or newer, and at least one supported coding agent: Claude Code, Codex, or OpenCode. Claude Code comes bundled with DorkOS, so it's the fastest way to start. One command installs DorkOS. No DorkOS account to create, no cloud services to configure. If you can run `npm install`, you're ready.
MIT (an open license with almost no restrictions). Use it commercially, fork it, modify it, ship it.
Yes. DorkOS runs wherever you put it: your laptop, a home server, or a cheap cloud box. Built-in tunnel support lets you reach it from anywhere.
DorkOS is free and open source. Running the agents themselves costs a little, since they call out to whichever AI company powers them; a night of work might run a few dollars. DorkOS doesn't add anything on top.
Your fleet is one step away.
Get started.
Terminal
~ $ curl -fsSL https://dorkos.ai/install | bash
Run in your terminal
Checks Node.js, installs via npm, offers setup wizard.