Your AI platform. On your infrastructure.

It runs where you run.

Install Eldric as a single signed package and it runs on your own hardware — on-premises, in your data centre, or fully air-gapped with no outbound calls. The same platform scales from a single box to a multi-node cluster; you add nodes, not cloud subscriptions. Your data and your models never leave your network.

How to install →

One install serves many models — and routes each request to the right one.

Eldric runs several models in parallel from a single install and sends each user or request to the model that should handle it. On multi-GPU machines it places models for you: a model per card, a large model split across cards, or many small models packed together.

• A handful of professional GPUs can host several large models, or many small ones, side by side.
• A larger GPU cluster scales by adding nodes — capacity grows with the hardware you add.
• Per-user and per-team personalisation: fine-tune adapters on a shared base model so teams get tailored behaviour without standing up a separate model each.

Talk to it from anything.

Eldric speaks the OpenAI-compatible API, so the client is your choice. Use the built-in web chat, or point OpenWebUI, your own application, an SDK, or an MCP client at the same endpoint. Eldric is the engine; you bring the front end you prefer.

Run models in-house, or front the ones you already have.

Eldric can serve models natively on your hardware, or sit in front of inference you already run — Ollama, vLLM, or a cloud API — and present all of them through one unified API. Migrate gradually, mix local and cloud, and give your clients a single endpoint regardless of what's behind it.

See how that compares to the alternatives →

Not just inference — the whole stack, through one API.

Beyond serving models, Eldric brings the pieces a real deployment needs, each reachable through the same API: multi-agent orchestration, retrieval over your own knowledge bases, intelligent routing, agent swarms, scientific data access, speech and media processing, and industrial device and IoT connectivity. One platform to run, one surface to integrate.

Browse the capabilities →

Bring your data, make the models yours.

You're not limited to off-the-shelf weights. Bring your own data and fine-tune with built-in methods — LoRA, QLoRA, DPO, distillation, and federated training that keeps each site's data local. Or skip training entirely and just serve the models we provide. Knowledge distillation can even compress a larger model's knowledge into a compact file that runs on CPU.

Less to install, less to break.

Eldric installs as a single signed package through your normal package manager — no dependency hell, no sprawl of services to wire together by hand. Updates are atomic, and configuration lives in the admin GUI rather than scattered config files.

Run it yourself.

Install the pre-built package — dnf install eldric-aios on RHEL / Fedora, or download the macOS installer — and browse to the chat shell. First signup becomes admin.

Get started Use cases  ·  Compare Eldric  ·  As a datacenter OS