CTRL NODE
Guides · Feb 28, 2026 · 5 min read

First Steps in CTRL NODE: Your First Remote Agent, Start to Finish

First Steps in CTRL NODE: Your First Remote Agent, Start to Finish

This is the shortest path from "never used CTRL NODE" to "watched an AI agent finish a real task on my own machine, from my browser." We'll use Claude Code as the example provider, but the same flow works with GitHub Copilot, Cursor, or Gemini — you just pick a different agent type in step one.

By the end you'll have: a project, a paired Bridge, one agent, and one completed task with real output files sitting in your local workspace.

1. Create a project

Sign in to CTRL NODE and the setup wizard walks you through your first project. Give it a name — we'll use Remote Claude Project — a short description, and, most importantly, the agent type.

Placeholder — new project wizard, name + agent type selection

The agent type is a project-level choice: every agent you add to this project will run through that same provider. If you want to mix Claude and Copilot agents, you'll create a second project for that.

Once you pick Claude, the wizard shows you the exact install command for Claude Code on your machine. Run it in your terminal — if it's your first time, it'll ask you to authenticate with your Claude account.

2. Pair the Bridge

Back in the wizard, there's a Bridge setup section with a download command for your OS. Grab it, run it.

Placeholder — wizard showing the Bridge download command

The first thing Bridge asks for is a workspace folder. This matters: agents can only read and write inside that folder, by design — nothing else on your machine is reachable, so there's no risk of an agent touching something it shouldn't. Pick a folder, and Bridge starts up and asks for a connection token.

Head back to the CTRL NODE wizard and hit GENERATE TOKEN.

Placeholder — GENERATE TOKEN button and the generated pairing token

Copy that token, paste it into the terminal where Bridge is waiting. That's it — Bridge is now paired to your account, and your browser can reach it.

3. Create your first agent

With the Bridge connected, create the agent that's actually going to run your tasks. Give it a name (CLAUDE AGENT works fine), a role, and — the important part — the specific Claude model it should use.

Placeholder — new agent form: name, role, model selector

There's an optional system instructions field too — anything you put there gets included in every prompt this agent receives, so it's a good place for standing context you don't want to retype every time.

4. Create and run a task

Time for a real task. Something simple but genuinely useful: generate a single-page site showing the current Bitcoin price with a 24-hour chart.

Generate an HTML page with inline CSS. Below the price, render a chart
of the Bitcoin price over the last 24 hours.
Placeholder — task creation form with the Bitcoin price prompt

Hit RUN. The task moves to In progress, and the Agent Activity panel starts filling in live — you can watch the agent look up the current price, draft the HTML, iterate. This isn't a spinner; it's the actual stream of what the agent is doing.

Placeholder — task in progress, live Agent Activity stream

Two panels are worth checking while you wait: Files shows your local workspace remotely — you'll notice the task's instructions are, underneath, just a markdown file sitting on your machine — and the eventual output folder, where the agent's generated files will land.

5. See the result

When the task finishes, the Result section summarizes what happened, and Files shows the output — in this case, an HTML page with the live Bitcoin price and a 24-hour chart. You can view it straight from the browser or download it.

Placeholder — completed task Result summary and generated output file

That's the full loop: project → paired Bridge → agent → task → result, all from the browser, with the actual work happening on your machine the whole time.

6. Where to go from here

A few sections worth knowing about once you've got the basics down:

  • Tasks — every task you've run, either as a Kanban board or a flat list, each with live status.
  • Team — manage every agent in the project: model, system instructions, one place for all of it.
  • Files — a remote browser for your workspace folder. View, edit, download, all without leaving the tab.
  • Dashboard — aggregate stats across your tasks and agents.

7. Bonus: reusing output as a template

One thing worth trying once you've got a completed task: use its output as a starting point for the next one. Take the Bitcoin HTML page from before, download it, and hand it back to the agent as a reference for a new task:

Following the design of the template I've attached, generate a new HTML
page showing the ETHEREUM price instead of Bitcoin.
Placeholder — new task referencing the previous output as a template, and the resulting Ethereum page

Hit RUN, watch the same Agent Activity stream, and a few moments later you've got a second page — same layout, different asset — without writing a single line of HTML yourself.

That's the whole first loop: a remote agent, connected through the Bridge to your own machine, running real tasks you can watch and inspect from anywhere. From here, every provider — Copilot, Gemini, Cursor, Codex, Hermes, OpenClaw — follows the same shape.