Skip to main content
Ontelis home
See a demo
Ontelis home

Run AI on your own hardware. We plan it, build it and run it.

We size the GPUs you need, install the inference hardware, models and data plane in your data centre, and keep them running. Your data and model calls stay inside your walls.

An open GPU server rack in an on-premises data centre
  • Your data stays in your building. Data, prompts and model calls stay on hardware you own. Air-gapped if your policy needs it.

  • You hold the keys. Keys live in your own key management service or hardware security module. We can’t unmask your data, even when we run the system for you.

  • Any open model, trained on your data. Run open-weight models like Llama, Qwen or Mistral, and fine-tune them on your own data. The weights stay yours.

  • Cost you can plan. A fixed platform you size once, instead of a per-token bill that grows with every new agent.

What we handle

  • Sized to your real demand

    We count your users, models and tokens, then size the GPUs, power and cooling to match. You see the monthly cost before you buy anything. If the cloud is cheaper for a workload, we tell you.

  • Built in your data centre, tested before go-live

    We check power and cooling, plan the racks and work with your hardware vendor on delivery. Nothing goes live until it passes 48 hours at full load and its first real inference.

  • Your data plane runs on your premises

    The data plane is where everything that touches your data runs. It sits on your premises, next to the open-weight models we install.

  • Only allow-listed metadata reaches us, and only if you connect

    A connected install sends opaque IDs, versions, counts, health signals and metering, through an allow-listed schema with no free-text fields. Air-gapped installs get offline, signed update bundles.

How it works

  1. Step 1: Map

    We list your AI use cases and mark which must run inside. Your security team and privacy officer sign off the boundary.

  2. Step 2: Size

    We turn users and models into a GPU count, a rack plan and a cost model you can check.

  3. Step 3: Build

    We order through your chosen vendor, install with your facilities team and test everything under full load.

  4. Step 4: Run

    Teams get access with quotas and an audit trail. Updates arrive as signed releases, in your upgrade window.

Who does the work

One team from the first workshop to handover, working with your IT, facilities and QA.

  • Solutions architect

    Sizes GPUs, network and storage to your workloads and builds the cost model.

  • Infrastructure engineer

    Checks power and cooling, plans racks and cabling, and oversees the install.

  • Inference engineer

    Picks and tunes the models for speed and cost per answer on your hardware.

  • Platform engineer

    Sets up access, quotas and monitoring so every team gets its share.

  • Validation engineer

    Writes and runs the IQ, OQ and PQ pack your QA team signs off.

  • Delivery lead

    Runs the plan end to end and keeps every date and owner visible.

Questions people ask

Ontelis On-Premises AI is a service: we help you set up and run your own AI inference, with your own models, on your own hardware. Our team sizes the GPUs to your real demand, installs the inference hardware and open-weight models in your data centre with your facilities team, and tests everything under full load before go-live. Your data, prompts and model calls run inside your walls, and the keys sit in your own HSM. After go-live we run it for you, or hand it to your team.

Where life sciences puts AI to work.

See a demo