The short version.
You have data that cannot be sent to an American cloud provider. Maybe it is patient records under HIPAA. Maybe it is privileged correspondence in a law firm. Maybe it is sensor data from a factory floor. Maybe it is research IP a competitor would love to read.
Today, you either avoid using AI for that work, or you ship your data to OpenAI and hope nobody at the other end takes a look. Eldric is the third option: the same modern AI capabilities, hosted on your own server, in your own building, with your own audit trail.
What is in the box.
- A chat shell for your browser. Looks like ChatGPT. Behaves like ChatGPT. Talks to your own AI.
- A native desktop client — macOS today, Windows and Linux in development. Works against your Eldric cluster, or directly against any cloud API you bring — the choice is per-session.
- A command-line client for scripts and terminals — same platform reach (macOS today, others in development).
- An OpenAI-compatible API, so tooling written for OpenAI works against your Eldric without modification.
- Knowledge-base search over your own documents, with the sources cited inline so you can verify any answer.
- Voice in, voice out for phone-style use without a keyboard.
- An AI-driven inbox across email, SMS, WhatsApp, Signal, Microsoft Teams, and XMPP — drafts proposed, human approves.
- A training pipeline to fine-tune a model on your own data, without uploading anywhere.
- 140 scientific-database integrations covering NASA, ESA, CERN, PubMed, NCBI, GBIF, IAEA, and many more.
- Industrial connectors for OPC-UA, Modbus, MQTT — for factory floors and IoT.
What is not in the box.
We do not require a U.S. cloud account. We do not charge per token. We do not phone home with telemetry by default. We do not host your data. We do not run the model for you. We are software; you bring the hardware.
Who it is for.
Eldric works for any organization that has both a) a real need for AI and b) a real reason its data cannot leave the building. In practice that is:
- Hospitals and clinical networks (HIPAA-aware).
- Banks, asset managers, and trading desks (data-residency and air-gap requirements).
- Universities and research institutes (IP protection, publication ethics).
- Pharma and biotech (GLP and FDA 21 CFR Part 11 compliance).
- Law firms (legal privilege).
- Insurance, public sector, and regulated utilities (sector-specific data laws).
- Factories, smart manufacturing, energy, and logistics (operational data sovereignty).
- Automotive, robotics, and field deployments (intermittent connectivity).
Who it is not for.
If you are building a public consumer app that lives entirely in someone else's cloud, you do not need Eldric — you can use OpenAI directly. If you have no regulated data, no IP to protect, and no compliance posture, the cloud is cheaper and easier. Eldric exists for everyone who falls outside that envelope.
What makes it different.
Three things, in order:
- Sovereignty. The architecture is shaped so that nothing critical depends on a third party. No "phone-home for license validation" in the data path. No hidden gateway. License signing is Ed25519, offline-verifiable.
- Honest scope. We name what works today and what does not. The home page lists the current numbers — including a known latency cliff in concurrent knowledge-base search that we have not yet fixed. We will not surprise you in production.
- Built in Vienna. Eldric is an Austrian product. We integrate the xLSTM architecture from the Linz lab of Sepp Hochreiter — the inventor of LSTM. We are independent of NXAI; we are part of the broader European AI community.
First run.
Once you install the packages and start the services, you open the chat interface in a browser. The first time the system sees a user sign up, that person becomes the administrator — they get tenant, user, license and configuration controls. There is no shipped default password; the admin is whoever signs up first on a fresh install. Make sure the right person on your team is the one in front of the screen, and lock the signup down right after.
Next step.
Two paths:
- If you want to try it, download a release and follow the quickstart. A single laptop will get you to private chat in fifteen minutes.
- If you want to talk it through first, send a note to office@eldric.ai. Tell us what you are trying to do; we will tell you whether Eldric is a fit and, if not, who else might be.