Neurix the connection
Neurix runs language models on your phone or desktop. Download a model once over Wi-Fi. Use it forever. After that first download it needs no internet, no account and no server.
Why I built it
I'm from Skardu, a small town in the mountains of northern Pakistan. It's a place where internet is a luxury, not a given. When you drive out toward Deosai or up the road to Shigar, the signal drops. Sometimes for hours, sometimes for days.
I relied on ChatGPT and Claude for everything: writing, coding, brainstorming, translating. But every time I left town, I lost access to all of it. That's why I built Neurix.
How it works
Open the app, pick a model from the store, tap download. Downloads can pause and resume, and each one is verified before use.
Once downloaded, the model runs on the device's own CPU. Conversations never leave it.
Chats are saved locally and can be searched. You can export them, along with your custom characters, and import them on another device.
The model manager warns when a model is too large for the device.
The models
There are 8, all GGUF files at Q4_K_M quantization.
| Model | Size |
|---|---|
| Qwen 2.5 0.5B | 380 MB |
| Llama 3.2 1B | 700 MB |
| Qwen 2.5 1.5B | 940 MB |
| SmolLM2 1.7B | 1.0 GB |
| Gemma 2 2B | 1.5 GB |
| Qwen 2.5 3B | 1.8 GB |
| Llama 3.2 3B | 2.0 GB |
| Phi-3.5 Mini | 2.2 GB |
How it is built
Neurix is a Tauri 2 app for Android and Windows. The Rust backend loads the model and runs inference through Candle, Hugging Face's ML framework for Rust. The interface is React and TypeScript.
Neurix is open source, MIT licensed.