Three install paths cover the great majority of deployments: a Linux server for a team or department, a macOS workstation for evaluation and solo work, or a multi-node cluster for a hospital network or research consortium. Pick the one that fits — the commands below assume root and a clean host.
The recommended path is a single bootstrap command that adds the signed repository, installs the metapackage, enables the service, and runs the post-install setup:
curl -fsSL https://repo.eldric.ai/install-eldric.sh | sudo bash
That bootstrap chains four steps: add the dnf repository, dnf install eldric-aios, systemctl enable --now eldric-aios, and a hand-off to eldric setup which waits for the daemon's health probe to come green. Within thirty seconds the controller, edge, router, worker and data daemons are all up. Open http://<your-host>:8880/chat in a browser — single-node listens on port 8880 by default. The very first signup becomes the system administrator. There is no default password. (When you put a dedicated edge gateway in front, the URL drops the port and becomes https://<your-host>/chat.)
Pass a license file and an admin email through the bootstrap by separating script arguments with -s --:
curl -fsSL https://repo.eldric.ai/install-eldric.sh | \
sudo bash -s -- --license-file /tmp/license.json --admin-email me@example.com
Prefer the steps separately — for example, to inspect the repo before installing, or to pin a specific version with dnf?
curl -fsSL https://repo.eldric.ai/install.sh | sudo bash sudo dnf install eldric-aios sudo systemctl enable --now eldric-aios sudo eldric setup
The first script only adds the signed dnf repository (its GPG signing key is pinned by fingerprint, so an intermediary serving a swapped key fails closed). eldric setup is the post-install bootstrap — it waits up to sixty seconds for the local daemon's /health probe, activates a license file if one was passed, checks whether an admin account already exists, and prints a one-screen install summary. It is idempotent; re-run it any time. Flags:
--license-file PATH — activate the signed license at PATH before printing the summary.--admin-email EMAIL — pre-fill the admin email in the summary.--dry-run — report what would happen without making changes.--quiet — suppress progress output; non-zero exit if anything failed.--json — emit the install summary as machine-readable JSON.The full install log is written to /var/log/eldric-install.log.
On macOS, pick the package that matches what you're doing: the desktop client for working against a remote Eldric cluster from your Mac, or the AI OS local daemon to run Eldric on your Mac itself.
The native SwiftUI app plus the eldric command-line tool. No daemons, no listening ports, no background services. Open the app, point it at your cluster (or at a cloud LLM provider directly), done.
Download Eldric desktop client
5.0.14 · 27 MB · sha256 02818d4d…439056e629
Installs /Applications/Eldric.app and /usr/local/bin/eldric.
Run Eldric itself on your Mac — single-node evaluation, plugin development, offline LLM workflows. Same architecture as the Linux server install, packaged for macOS with every dependency bundled (no Homebrew required at runtime).
Download Eldric AI OS for macOS
5.0.0 · 16 MB · sha256 53eb2fe9…ca0057ce0b
The installer places the eldric-aios binary under /usr/local/libexec/eldric-aios/ and offers a choice between two launch modes:
127.0.0.1. Install and uninstall without sudo. Best for solo evaluation._eldric user, bound to 0.0.0.0. Mirrors the Linux server install; requires sudo to remove.After install the chat interface comes up at http://127.0.0.1:8880/chat. The first signup on that machine becomes the local administrator.
Gatekeeper note. macOS will prompt “unidentified developer” on first launch of either package. Right-click the PKG, choose Open, then confirm with Open again in the security dialog. Eldric 5.0 packages are not Apple-notarized; notarized releases are planned for an upcoming 5.0.x patch.
For a hospital network, research consortium, or any deployment that needs more than one node, install the controller on one machine first, then join workers, routers and data nodes from there. The cluster setup guide walks through SSH-based fanout deployment, license placement, replication policy and TLS configuration.
Web chat at /chat, twenty-five specialised agents (researcher, coder, validator, planner, scientist, and more), tool-calling out of the box.
NASA, ESA, CERN, LIGO, ClinicalTrials.gov, PubMed, BLAST, AlphaFold, Materials Project, and the rest — all federated through the science worker.
Browse and install plugins from /api/v1/marketplace/catalog. Five plugin types: Tool, Filter, Pipe, Action, Widget.
Departments, studies or projects each get their own tenant. Viewer, Developer, Admin, SuperAdmin roles. Cross-tenant access denied at the gateway.
Fine-tune a model on your own data — LoRA, QLoRA, SFT, DPO. The dataset never leaves the data worker.
Hash-chained log of every AI-assisted decision. Defensible record for compliance reviews.
The free tier ships with the full kernel: one controller, one router, two workers, all features. To scale beyond that, request a Standard, Professional or Enterprise license from license@core.at and drop the signed license file into /etc/eldric/license.json. The controller picks it up on next start and the limits lift accordingly.
Installed, but nothing's happening? Walk through the first-run guide. Something failed during install? Common errors and recoveries are collected in the troubleshooting page. Stuck on something neither covers? Write to office@eldric.ai.