LexiSuite Docs

Getting Started

From download to a running model in five steps. You need a PC with a 64-bit UEFI firmware, a USB stick of 8 GB or more (everything on it is erased), and a second device with a browser. A GPU is optional but is the point: AMD, Intel and NVIDIA cards are used through Vulkan, NVIDIA also through CUDA.

1. Download the newest image

The newest image, 5db82d6 (November 13, 2026), is on the download page; huggingface.co/Wa1k3r/LexiSuite has the images before it. Earlier images are on the Releases page; there, take the newest release. It has three kinds of file:

File What it is
LexiSuite-<id>-secureboot-dev.img.xz The USB image (about 350 MB compressed, 3 GiB written). <id> is the source commit it was built from.
LexiSuite-<id>-secureboot-dev.img.xz.sha256 Its checksum. Keep it in the same folder as the image.
install-to-usb.sh The Linux installer script (when the release includes it).

The image contains LexiOS on Linux 7.4 and LexiKernel, the Panel and LexiEngine. It contains no models; you add your own in step 5. What changed in each image is on Release Notes.

2. Write it to a USB stick

Linux, one command. With the image, its .sha256 file and install-to-usb.sh in one folder:

sudo bash install-to-usb.sh

The script checks the image against its checksum, lists only USB disks, refuses a disk that is in use or too small, makes you type the disk's name, writes the image, reads it back to compare, and ejects. bash install-to-usb.sh --dry-run runs every check and writes nothing.

Windows or macOS. Use balenaEtcher or Raspberry Pi Imager and give it the .img.xz file directly. Details and the manual dd method are on Install from USB.

3. Boot it

Restart the PC, open its boot-device menu (often F12, F11 or Esc) and choose the USB stick.

The boot menu offers two systems. Start with 2 LexiOS on Linux 7.4: it supports the widest range of hardware and runs the GPU natively. 1 LexiKernel is LexiSuite's own Rust kernel; it needs an active IOMMU (Intel VT-d or AMD-Vi) and is still limited. Kernels compares them.

On the first boot the data partition grows to fill the stick.

4. Open the Panel and create the administrator

The machine's own screen shows a status page with the Panel address (port 8090) and, when one is required, a setup code (Lp1- and four groups of four; type it in either case, with or without the dashes). Open that address from a browser on the same network, enter the setup code if the page asks for one, and set the administrator and root passwords. Do this straight away and on a network you trust: until it is done, the setup page is open. More on First Boot.

5. Add a model and chat

The Panel has tabs across the top: Status, Models, Run, Optimize, Fleet, Files, System, Help. Each opens a menu of its pages.

  1. Models: download a GGUF model from the library, or upload one you have.
  2. Run: choose the model, the engine (LexiEngine) and the GPU for an instance and start it; Instances has the exact steps. The engine sizes its memory from the card; there is no budget to set.
  3. Status → Chat: pick the running model and send a prompt. Any OpenAI-compatible or Anthropic-compatible client can use the same instance; see OpenAI-compatible API.

Help → Find a feature searches every page and setting by name, and Help → Guided tasks walks through jobs such as "serve your first model" step by step.

If something does not start, the Panel names the reason; Troubleshooting lists the common ones. For speed comparisons use the method on Benchmarks.

Next: Hardware · Panel Overview · FAQ · Licence.