Overview

Get started

Install Eldric on hardware you control, then browse to the chat. Pre-built packages — no build from source, no cloud account, no GPU required.

Requirements

Fedora 42+, RHEL 9+, CentOS Stream 9+, or a RHEL-family derivative (Rocky, Alma), on x86_64.
A CPU is enough — a GPU only makes inference faster. It runs from a single box up to a cluster.
Root (or sudo) on the target host.

Install

The bootstrap script adds the repo, installs the package and enables the service:

curl -fsSL https://repo.eldric.ai/install-eldric.sh | sudo bash

Or add the repo and install the package yourself:

curl -fsSL https://repo.eldric.ai/install.sh | sudo bash
sudo dnf install eldric-aios

GPU hosts On a GPU node, install eldric-aios-cuda for CUDA acceleration. The CPU package runs everywhere; the CUDA package adds GPU inference.

At the end of the bootstrap script The script finishes by handing off to the first-run step, which in this release has to be run interactively. You will see it stop with a message about non-interactive setup. The repo, the package and the service are installed and running at that point — carry on with First run below.

First run

The first-run step is interactive. It asks for an admin email and password, a license — the path to your enterprise_license.json, or free for the free tier — and a default tenant name:

sudo eldric setup --interactive

The license you give it is installed where the service looks for it. Restart to apply:

sudo systemctl restart eldric-aios

Then browse to https://<host>/chat and sign up. The first account to sign up becomes the admin — that signup, not the first-run step, is what creates your login.

Next steps

Call the API — the OpenAI-compatible endpoint plus the /api/v1 surface.
Use RAG — point Eldric at your own documents.
Operate a cluster — tenants, roles, storage, upgrades.
Troubleshooting — if something doesn't come up.