███╗   ██╗ ███████╗ ██╗  ██╗ ██╗ ██╗
 ████╗  ██║ ██╔════╝ ╚██╗██╔╝ ██║ ██║
 ██╔██╗ ██║ █████╗    ╚███╔╝  ██║ ██║
 ██║╚██╗██║ ██╔══╝    ██╔██╗  ██║ ██║
 ██║ ╚████║ ███████╗ ██╔╝ ██╗ ██║ ███████╗
 ╚═╝  ╚═══╝ ╚══════╝ ╚═╝  ╚═╝ ╚═╝ ╚══════╝

A personal assistant that runs locally on your machine with a familiar CLI interface.

Reason for project existence

(Skip if you want, not necessary to read)

My laptop has a NPU and it does nothing. So, I did some research and I thought it would be a cool project to run a local llm that can do stuff locally (tool callls and what not), espcially with OpenClaw being a big hype rn, I decided to do it... and this is the result of it. Since, I have an Intel processor I went with OpenVino as the base (I plan to support all hardware and brands). But, from what I know this is the only way to run stuff of the NPU. Some people said stuff about Onnx runtime would work but I dont know. I have a lot of ideas to make this tool more use full. As of now it only tells time (lmao). I feel this project has a lot of potential. Let's see where this goes.

2026-03-06.21-10-33.mp4

Features

• Run LLMs locally on Intel NPU and GPU. (GPU not supported as of now but could theoritically work).
• Native tool calling (Ex: date/time lookups, only tool as of now)
• Rich terminal UI with streaming responses
• Download and manage OpenVINO-optimized models
• Configurable via TOML

Requirements

• Linux (tested on Arch based system, should work on any distro)
• Python 3.12+
• Intel Core Ultra processor (Meteor Lake / Lunar Lake) for NPU support
• Intel NPU driver
• OpenVINO runtime (installed automatically via pip)

NPU driver setup

The Intel NPU driver is required to run models on the NPU. Without it, you can still run on CPU.

Arch Linux (AUR):

yay -S intel-npu-driver

or

paru -S intel-npu-driver

Ubuntu/Debian:

Download the latest .deb packages from the intel-npu-driver releases and install:

sudo dpkg -i intel-driver-compiler-npu_*.deb intel-fw-npu_*.deb intel-level-zero-npu_*.deb

Fedora:

Download the latest .rpm packages from the intel-npu-driver releases and install:

sudo rpm -i intel-driver-compiler-npu-*.rpm intel-fw-npu-*.rpm intel-level-zero-npu-*.rpm

After installing the driver, add your user to the render group to access the NPU device:

sudo usermod -aG render $USER

Log out and back in for the group change to take effect.

You can verify the NPU is available by running:

nexil devices

Installation

From PyPI:

pip install nexil

From GitHub:

pip install git+https://github.com/James-Crockett/nexil.git

Quick start

Download a model:

nexil download --model-id Qwen/Qwen3-4B

Thinking models work better with this application, so I'd recommed it. Instruct models do work but not well with the tools.

Start the assistant:

nexil

Or launch the CLI chat directly:

nexil chat

Commands

Command	Description
nexil	Start the chat assistant
nexil chat	Start CLI chat mode
nexil download	Download a model
nexil models	List installed models
nexil devices	List available devices

Chat commands

Command	Description
/help	Show available commands
/model	Switch model
/device	Switch device
/clear	Clear chat history
/quit	Exit the assistant

Configuration

Config file: ~/.config/nexil/config.toml

model_path = "/path/to/model"
device = "NPU"

NPU Monitoring Tool:

https://github.com/DMontgomery40/intel-npu-top

License

Apache 2.0 — see LICENSE for details.