tool.002 — runs on your machine, stays on your machine
GobboNet
A local front end that respects your time and your privacy. Mac and Linux versions coming soon.
↓ Download for Windowseverything GobboNet sends, ever
| first run | github.com | downloads the engine that runs the model | once |
| first run | huggingface.co | downloads the model you picked | once |
| every message | — | nothing | 0 bytes |
| every chat | — | nothing | 0 bytes |
| from then on | — | nothing | 0 bytes |
what it is
A local front end that respects your time and your privacy.
Most AI chat happens on somebody else's computer. You type, it goes to a company, the company answers, and the company keeps a copy. When it does happen locally, systems are often kept behind technical barriers that can keep new folks from engaging.
GobboNet flips that. One download, double clicked, and everything runs. The model file sits on your hard drive. The program answering you is running on your own graphics card. Your conversations are saved in your browser, on your machine, and there is no server anywhere to send them to.
That means no account, no API key, no monthly bill, and no terms of service that change under you. It also means nobody is reading your chats to train anything, because there is no anybody in the loop except YOU.
what you get
What's in the box
Nothing is collected
No telemetry, no crash reports, no usage stats, no "anonymous" analytics. There is no collection code to switch off.
Works with no internet
Once the model is downloaded you can pull the plug. It keeps answering. Handy on a plane, essential if you don't trust the network.
No setup ritual
No Python, no Docker, no command line, no config files. Run the installer, click the shortcut, pick a model from a list.
Picks a model that fits
It checks what graphics card you have and only offers models your machine can actually run. AMD, NVIDIA and Intel all work.
Reach it from the couch
Optional: open it to your home network and use it from your phone. Local network only, password protected, never exposed to the internet.
Readable all the way down
The whole thing is scripts and a web page. Open the install folder and read it. No compiled black box doing something you can't inspect.
getting started
Three steps, once
-
Download and run the installer
It's 681 KB and takes a couple of seconds. Installs to your own user folder, so it never asks for administrator rights.
-
Open GobboNet and pick a model
First launch fetches the engine and shows you a list of models that suit your graphics card. Pick one and it downloads. This is the slow part — anywhere from a few minutes to an hour depending on what you choose.
-
Start typing
The chat UI opens in your browser. Everything from here runs on your machine. Later launches skip straight to this step.
before you click
Things worth knowing
Windows will warn you about it
You'll see a blue "Windows protected your PC" box. Click More info, then Run anyway. This happens because the installer isn't signed with a paid certificate — the kind that costs a few hundred a year and mostly proves you have a few hundred a year.
Models are big
The installer is tiny, but the AI model itself isn't. Small ones are a couple of gigabytes; large ones run to forty. GobboNet shows you the size before you commit to a download, and only lists ones your machine can run.
Windows only, for now
Linux and Mac fork coming soon, planning docs have been drafted but we want to ensure the recent refactor is bug free before we begin this split. GobboNet is built out of Windows scripts, so that's where it runs. Windows 10 and 11, 64-bit. A dedicated graphics card makes it much faster, but it will fall back to your processor if you don't have one.
the source