Organizing the world

/ɔrdnə/ — to arrange, fix, take care of
to organise, put in order

Every agent and every server your project needs, in one window

Run Claude Code, Codex and Gemini CLI beside the dev servers, workers and watchers they are changing — one workspace per project folder, with an MCP toolbelt so your agents can read the output, restart what crashed, and hand work to each other.

Free for everything on one machine.

Running agents broke the terminal

One agent was fine. Three agents, a dev server, a queue worker and a log tail is a different job, and a grid of tabs is not the tool for it.

Which one is waiting on you?

Three agents running, nine tabs open, and the only way to find the one that asked a question is to click through all of them.

Agents working blind

Your agent changed the API and the dev server died on reload. It has no way to look, so it carries on as though the change worked.

It crashed twenty minutes ago

The queue worker exited, everything downstream has been failing since, and nothing told you. Its tab is four to the left.

Open a folder, get a workspace

Open the folder and the workspace is already assembled: the agents, the processes that serve the app, and the shells you poke at it with — in one window, in one rail, all visible at once.

The whole project starts at once

Dev servers, workers and watchers come up with the workspace, from a config file you commit. Supervised, restarted when they crash, restarted when a file changes.

You can see who is working and who is stuck

Every agent runs in a real PTY with idle detection, so its row tells you whether it is thinking or waiting on you — without opening it.

Your agents can read the stack they are changing

An MCP server over the workspace: read process output, start and stop things, wait for a port to bind, take a lock, file a todo. The agent can check whether its own change worked.

Agents hand work to each other

Messaging between agents is a first-class primitive rather than coordination smuggled through a shared file, so one agent can pass a task to another and know it arrived.

Reach a project on a machine you are not sitting at

A project on one machine is attached to another, and the attachment is the grant: you pair a machine for a project, and that is the only thing it can reach. From then on it behaves like any local project — same rail, same output, and an agent in one can hand work to an agent in the other.

01 — Two machines

Your work computer keeps running. Anything else just connects to it.

1

P1 and P2 stay running on your work computer.

2

Open Ordne on a second desktop — it has its own projects.

3

Your work machine’s projects show up right beside them.

02 — Any screen

The same projects, in a browser or in your pocket.

1

Ordne runs on your desktop, hosting P1 and P2.

2

Open the identical view in a remote browser.

3

Or on your phone. Nothing to install, nothing to sync.

Work computer
ordne — local
Projects · running here
P1:3000
P2:4200
host · listening on :7000
Second desktop
ordne — local + remote
On this machine
P3:3000
P4idle
From work computer
P1remote
attached
Desktop app · the host
ordne — local
Projects · running here
P1:3000
P2:4200
host · sharing 2 projects
Remote browser
ordne.local · yourname
Projects
P1remote
P2remote
ordne
Projects
P1↗
P2↗
Phone

Local is free. Remote is paid.

The free tier is complete rather than a trial — no limit on projects, processes, agents or terminals, and nothing metered. You pay when you want to reach a machine that is not the one in front of you.

Free

$0
  • Unlimited projects, processes, agents and terminals
  • The full MCP toolbelt and the coordination tools
  • Every feature on the machine you are sitting at
  • No project or process caps, ever

Your code never leaves your machine

Agents run locally, against your working copy, in your shell. No cloud sandbox, no repository upload, no copy of your project on our side — because there is no our side to keep it on. The hosted service knows your email address and your plan.

Questions

Which agents does it run?

Anything that runs in a terminal — Claude Code, Codex, Gemini CLI, Amp, OpenCode, Copilot CLI, aider. There is no wrapper and no lock-in: each one gets a real PTY, the same as it would get from your shell.

Do you see my code?

No. Projects, source, terminal output, scrollback, todos and scratchpads live in local storage on your machine. The hosted side holds your account and which plan you are on — not your work, and not so much as a directory path.

Why does a local tool need an account?

So that paid remote access is enforced by a service rather than by a flag in the app that anyone could flip. You sign in once; local features never consult it again, and never fail because the network did.

What does free actually include?

Everything on the machine in front of you, with no caps — projects, processes, agents, terminals, the full MCP toolbelt and the coordination tools. Nothing is metered and nothing counts down.

Does the other machine have to be running?

Yes. A phone or a browser is a client, not a host — it attaches to a desktop that is awake and running your processes. When that machine is off or off the network, the project shows as unreachable, which is a state of its own rather than a pane that looks empty. When it comes back, output catches up from where you left it.

Is this another IDE?

No. It is a terminal workspace, and your editor stays your editor. Nothing here asks you to move where you write code.