Product privacy

Private by design. Easy to check yourself.

“How do you know it’s private?” is the right question to ask about any AI tool. Here is how the drive keeps your work on your computer, what we measured, and five checks you can run yourself without taking our word for it.

Check it yourself See what we measured

Requests stay on the computer itself
Passes the airplane-mode test
Optional passphrase encryption for saved work

The dashboard’s Privacy and storage panel. It lists the AI server at http://127.0.0.1:8080 and reports 0 external or nonlocal requests blocked. — Measured on our development computer while the dashboard wrote a set of practice questions. You can run the same checks yourself.

The short answer

Your work stays on your computer, because nothing in the software sends it anywhere

Cloud AI tools send what you type to a company’s servers, where the AI runs. The Evolve AI drive brings the AI to you. The AI model and the program that runs it sit on the drive and run on your own computer’s processor and memory.

When you ask the dashboard for a lesson plan, your request travels from your browser to that program over an address that exists only inside your computer: 127.0.0.1. No other computer can reach that address, and the program is set up to answer nothing else.

That is why the drive keeps working with Wi-Fi switched off. An AI that depended on the internet would stop.

The dashboard’s Privacy and storage panel. The AI server row reads http://127.0.0.1:8080, a loopback address on this computer. The Dashboard requests row reads: 2 local requests observed, 0 external or nonlocal requests blocked. — The dashboard’s built-in Privacy & storage panel reports, live, where every request goes.

Evidence

What we measured

We recorded the network activity of the Educator Edition while it wrote a set of practice questions on our development computer (Windows 11, version 0.8.0, October 6, 2026).

24 of 24

network requests from the dashboard went to 127.0.0.1:8080, the AI program on the same computer

requests went to any other address on the internet or the local network

127.0.0.1 only

the only address the AI program listened on, so other computers had no way in

How we checked

Chrome’s own network log recorded every request the dashboard made while generating the draft.
Windows reported every network connection owned by the AI program, llama-server.exe. All of them were between the browser and the program on the same computer, and it had no other open connections.
The launcher starts the program with the setting --host 127.0.0.1, which restricts it to this computer.

These results come from one computer. The checks below let you see the same thing on yours.

PS> Get-NetTCPConnection -OwningProcess (Get-Process llama-server).Id |
      Format-Table LocalAddress,LocalPort,RemoteAddress,RemotePort,State

LocalAddress LocalPort RemoteAddress RemotePort    State
------------ --------- ------------- ----------    -----
127.0.0.1         8080 127.0.0.1          56248 FinWait2
127.0.0.1         8080 0.0.0.0                0   Listen

Actual output on the development computer. The AI program listens only on 127.0.0.1, and its one connection is with the browser on the same computer. (0.0.0.0 on the listening line is a placeholder meaning “waiting for a connection.”)

Check it yourself

Five checks you can run yourself

Start with the first one. It takes two minutes and needs no technical knowledge.

The airplane-mode test

Turn on Airplane mode in Windows (or switch off Wi-Fi and unplug any network cable). Start the drive, open the dashboard, and create a draft.

What you will see: the draft is written as usual. An AI that relied on an online service could not answer you with the internet switched off.
Ask the dashboard

Select Privacy & storage at the top of the dashboard.

What you will see: the AI server address http://127.0.0.1:8080 and a count of the dashboard’s requests, with the number of outside requests it blocked.
Watch the browser’s network log

In Chrome or Edge, press F12, choose the Network tab, then create a draft.

What you will see: every request in the list goes to 127.0.0.1:8080.
Ask Windows who the AI program talks to

Open PowerShell while the AI is running and paste this command:
```
Get-NetTCPConnection -OwningProcess (Get-Process llama-server).Id
```
What you will see: every address listed is 127.0.0.1. You can also open Resource Monitor, choose the Network tab, and find llama-server.exe under TCP Connections.
For IT staff: inspect and block it

The AI program is llama.cpp, open-source software whose code anyone can read. The model, Qwen3-4B, is openly published. The dashboard is plain HTML and JavaScript files on the drive, and its security policy limits it to its own files.

IT staff can also add a Windows Firewall rule that blocks internet access for llama-server.exe, then create a draft. The program only uses connections inside the computer, so it has no need for that access.

An honest distinction

Private and secure: the difference

Private describes what the software does with your work. The Educator Edition keeps it on your computer during processing, and the checks above let you confirm that.

Secure also depends on everything around the software: the computer’s health, its other programs and browser extensions, who can pick up the drive or the laptop, and where you save exported files. No product can promise that for you.

So we make a narrower promise that you can test: this software sends your work nowhere. Then we give you tools for the rest: passphrase encryption for saved workspaces, controls for what each request includes, and buttons that clear your session.

Comparison

Compared with a typical cloud AI service

Where your work goes with each approach
Question	Evolve AI drive	Typical cloud AI service
Where the AI runs	On your computer	On the provider’s servers
Where your text travels	From your browser to a program on the same computer	Over the internet to the provider
Whose policies govern your text	Yours and your school’s	The provider’s terms and settings
Works with the internet off	Yes, after setup	No
Account required	No	Usually
Can you verify where requests go	Yes, with the checks above	You rely on the provider’s statements

Local processing

How requests stay local

The supplied dashboard sends AI requests to a server running on the same computer.
Its supplied configuration binds that server to the local loopback address, 127.0.0.1, which only the same computer can reach.
The dashboard uses no cloud AI services, no external analytics, and no internet fonts.
Document processing and optional English OCR (optical character recognition) run locally with components bundled on the drive.
The AI has no built-in web browsing, so it cannot look up websites or send your information to them.

The dashboard also includes a request guard that refuses non-local addresses, and a content security policy that limits it to its own bundled files.

Offline use

Working offline

After the required software and models are installed, the core workflows can run with networking disconnected. To work fully offline, turn off Wi-Fi and unplug any network cable before you start.

A connection may still be needed for first-time setup, for receiving updates, and for downloading software or models.

Controls

Your privacy controls

The dashboard starts in a temporary session and saves nothing unless you ask it to.

Preview requests

See the exact system and user messages the dashboard will send to the local model before you generate.

Choose what is included

Switch course settings and imported source text on or off for future requests.

Clear imported documents

Remove imported text and pending packet inputs from the session.

Clear drafts and sessions

Clear the current draft and undo history, grading results, or the whole session’s courses, materials, and form values.

Save only when you choose

Workspace files are written only when you select Save workspace or download a backup.

Optional encryption

Protect saved workspace files with a passphrase using AES-256-GCM encryption, with a key derived through PBKDF2-SHA256 at 600,000 iterations.

Diagnostics downloads contain metadata only, such as counts and statuses, so they can be shared for support without your document text.

Educator Edition

Grading data storage

Uploaded papers, rubric data, proposed scores, and approvals are temporary session data held in browser memory.
The grading batch is excluded from course workspace backups.
A grading report you deliberately keep as a course draft is included in workspace files you save afterward.
Exported reports and CSV grade sheets are separate files. They stay where your browser saved them until you delete them.

Limits

Limits you should know

Local processing narrows where your information travels, and some risks remain outside its reach.

Website privacy

This website is separate from the product

Everything above describes the dashboard on the drive. This public website runs on an internet web host. When you submit an order request, compatibility question, or contact form here, your answers travel over the internet and are stored in our website database so we can reply.

Please never upload or paste student papers, grades, or other student information into this website. The site has no feature for receiving them.

Read the website privacy notice

Bring private AI into your teaching

The Educator Edition is available now on three drives, starting at the Pocket option. Choose your drive, or check that your Windows computer can run it first.

See pricing Check compatibility