Ntense beyond the cloud: Use your own browser, development tools, and GPU

You do not have to move all your work into a Ntense Cloud Workspace. Ntense Tunnel connects a computer you already use to your Workspace, so Ntense can work with your Chrome browser, run commands where your tools are installed, and show you a website running on that machine. You choose how much access to give, and you can stop or revoke it at any time.

Tunnel is one of three ways to work with Ntense. Pick the one that matches what you want Ntense to do:

• Describe an app or website and try a working version without coding — Create an app from a prompt with Ntense
• Call Ntense models from your own app, script, or notebook — Use the Ntense AI API
• Let Ntense build, test, or troubleshoot with tools on another machine — connect that machine with Ntense Tunnel (this guide)

What can Ntense Tunnel connect?

Your Chrome browser. Some tasks need a real browser: checking a page you are signed in to, or testing a site the way a user sees it. Browser-only mode shares your existing Chrome with the Workspace so you can watch what Ntense does. It cannot run commands or share websites from your computer. First you turn on remote debugging in Chrome, and Chrome may ask you to approve the connection.

Your development computer. Your Mac may have Xcode, your laptop may have Android tooling, or your desktop may have a GPU setup you have already configured. With full control mode, Ntense can build, run, and debug work on that machine, using your project files where they already are. Connecting a tunnel does not install anything for the task. The tools, devices, and permissions a task needs must already be on that computer. By default, every command waits for you to approve it on your computer.

A notebook or other cloud runtime. A Jupyter, Google Colab, or Kaggle notebook on a Linux x86-64 runtime can also be a command target, if the service lets it make outbound network connections. This can give you a ready-made environment or a GPU you can use for learning. Notebook mode works differently: commands run without local approval, and Chrome stays on your own computer. More on this below.

In every mode, the tunnel makes an outbound connection from your machine to your Workspace. You do not need to open an inbound port or set up SSH to use it.

Websites on the connected machine. When command control is connected, Ntense can start a development server on that machine and connect its port to your Workspace. The site then gets its own Workspace web address, shown under Connections as Open website, so you can test it in any browser. Nothing is shared automatically. Each site is connected only when you or Ntense set it up. Your app handles its own sign-in, so do not leave private data behind an unprotected development site.

Which connection mode should I choose?

The Tunnel & SSH → Ntense Tunnel page in your Workspace gives you a start command for each mode. Choose the one with the least access that still gets the task done.

• Local browser use only: run it on the computer with your Chrome. It shares Chrome only and allows no commands. Stop it with Ctrl+C.
• Local full control: run it on the computer you want Ntense to work on. At startup you choose Chrome, commands, or both. Commands ask for your approval by default. Stop it with Ctrl+C.
• Jupyter / Colab / Kaggle: run it in a Linux x86-64 notebook runtime. It shares commands, plus websites running in the runtime. Commands run without approval. Stop it with a stop cell.

How do I start safely?

1. Generate a tunnel token

In your Workspace, open Tunnel & SSH → Ntense Tunnel and select Generate tunnel token. The token is shown only once, so copy it straight into your start command. Treat it like a password. If you generate a new token or select Revoke tunnel access, every tunnel using the old token disconnects.

2. Install Ntense Tunnel

Install on each computer you want to connect. It installs into your user folder, does not change your PATH or shell profile, and asks before it replaces an existing installation. Supported platforms are macOS on Apple silicon, Linux x86-64, and Windows x86-64.

macOS or Linux (shell):

curl -fsSL https://ntense.ai/ntense-tunnel-install.sh | bash

Windows (PowerShell):

Invoke-WebRequest https://ntense.ai/ntense-tunnel-install.ps1 -OutFile ntense-tunnel-install.ps1; & .\ntense-tunnel-install.ps1

3. Start the mode you need

Copy the start command from your Workspace's Tunnel Setup page. The page fills in your Workspace host and the token you just generated. The examples below use placeholders.

Local browser use only. Open chrome://inspect/#remote-debugging in Chrome, turn on remote debugging, and then run:

bash ~/.local/share/ntense-tunnel/ntense-tunnel.sh --browser-only --host '<WORKSPACE_HOST>' --token '<TUNNEL_TOKEN>'

Local full control. At startup, the tunnel asks whether to share Chrome and whether to allow commands. Both default to No.

bash ~/.local/share/ntense-tunnel/ntense-tunnel.sh --host '<WORKSPACE_HOST>' --token '<TUNNEL_TOKEN>'

On Windows, run the same options with the PowerShell launcher:

& "$env:LOCALAPPDATA\nTense\Tunnel\ntense-tunnel.ps1" --host '<WORKSPACE_HOST>' --token '<TUNNEL_TOKEN>'

When Ntense asks to run a command, your terminal shows the exact request. Press y to allow it once. Press a to allow commands with the same start, such as every git status, until the tunnel stops. Press n to reject it. Approving an interactive shell or coding agent gives control of that whole session, so use y and a with care. A prompt you do not answer in two minutes is rejected.

The Workspace page shows what is connected: websites, Chrome, commands, and the computer's operating system and approval mode.

Can I use Google Colab or Kaggle instead of setting up a GPU?

Yes, in two different ways.

To call the Ntense AI API. A Colab or Kaggle notebook with internet access can call the Ntense AI API from code, using your own API key. Follow the API setup guide and keep the key in the notebook provider's secret storage, not in a notebook cell. This lets you try an AI feature without installing a development environment. See the Google Colab FAQ^[1] and Kaggle Notebooks documentation^[2].

To let Ntense work inside the notebook runtime. Use the Jupyter / Colab / Kaggle tab on the Tunnel Setup page. Run the install in a notebook cell:

!curl -fsSL https://ntense.ai/ntense-tunnel-install.sh -o /tmp/ntense-tunnel-install.sh
!bash /tmp/ntense-tunnel-install.sh --tunnel-only

Then start it in the same runtime:

!bash ~/.local/share/ntense-tunnel/ntense-tunnel.sh --tunnel-only --host '<WORKSPACE_HOST>' --token '<TUNNEL_TOKEN>'

Notebook mode has no terminal for approval prompts, so commands run without asking you. Treat the runtime as something Ntense can fully control. Do not connect a notebook that holds data or credentials you would not hand over. Keep the token out of notebook cells you share and out of saved outputs.

If you also want Ntense to use your browser, run browser-only mode on your own computer with the same Workspace host and token. Commands then run in the notebook, and Chrome stays on your computer.

To stop the tunnel, run this in a later cell:

!bash ~/.local/share/ntense-tunnel/ntense-tunnel.sh --stop-all

Stopping the tunnel leaves the notebook kernel running. If the runtime resets, run the install and start cells again, and check which files and tools are still there before you continue.

Notebook providers set their own rules for remote shells, GPU use, and session length. These rules differ between providers and can change. Read the provider's terms before connecting a tunnel, and check that the runtime has the tools your task needs. Free resources can save setup time and money, but they may not always be available. See the Google Colab FAQ^[1] and Kaggle Notebooks documentation^[2].

When you are done

Press Ctrl+C in the terminal running the local tunnel, or run the stop cell in a notebook. If you think the token has been exposed, select Revoke tunnel access in your Workspace. Every connection using that token disconnects.

The learning loop is the same wherever the work runs: choose a real outcome, let AI help build it, inspect what happened, test the result, and learn the next concept the work brings up. If you want to see a working result first, start with an app from a prompt. When you are ready to build more yourself, move to the API or a connected development machine.

Sources and further reading

• Google Colab FAQ^[1]
• Kaggle Notebooks Documentation^[2]