A future that almost arrived.
The year Windows XP launched. The year Kubrick’s monolith first whispered to mankind. Two icons of a tomorrow that never quite came — until now.
Modern AI.
Ancient hardware.
NeuralXP runs a real, offline language model on machines the rest of the world abandoned twenty years ago. No cloud. No subscription. No telemetry. Just intelligence, where there used to be silence.
Tiny. Fast. Forever yours.
Old machine.
New mind.
Lightning fast
Optimized inference engine. No bloated frameworks. NeuralXP boots faster than your Pentium III ever booted Windows.
Runs on potatoes
512MB RAM. Single-core CPU. No GPU required. If it can run XP, it can run NeuralXP. Period.
Fully offline
Zero network calls. Zero telemetry. Zero accounts. Your prompts never leave the machine they were typed on.
Built to last
No forced updates. No deprecated APIs. No model retirements. The version you install today still works in 2046.
Native Windows app
Real .exe installer. Real Windows UI. No browser tabs, no Electron, no 800MB Chromium overhead. Just a program.
Open source
Every line on GitHub. GPL v3. Audit it, fork it, run it on a Windows 98 ThinkPad in an attic. We’d be honored.
“Open the model
files, HAL.”
The cloud took your data. The cloud took your privacy. The cloud took the AI itself and locked it behind monthly fees. NeuralXP takes it back — and runs it on a machine that predates the cloud entirely.
A real Windows XP application.
Native UI. Real installer. Powered by Qwen2.5-0.5B via llama.cpp — running entirely on the local machine. No browser. No Electron. No cloud.
Free. Forever.
NeuralXP is open-source software under the GNU General Public License v3. No accounts, no servers, no compromises — just AI you can hold in your hands.