Home / Installation
Private AI installation on your own servers
We can scope a deployment of the model on hardware you own or rent, behind your own firewall, connected to your existing software. We have not yet delivered an on-site installation, so the first ones will be scoped and tested closely with the client.
What it is
Installation is the work of turning a trained model into a service your staff and software can use. We size the hardware, set up the serving software, secure it and wire it into your systems. The model runs inside your network, so prompts, documents and logs stay where you already keep sensitive data.
It is meant for a model we distilled for you or an open-weight model you already have.
Who it is for
Organizations with data they cannot send to an outside provider, whether for contract, regulatory or plain business reasons, and IT teams that want a clear picture of what is running on their hardware and who can reach it.
What you would get
- A working deployment. The model served on your servers or private cloud, with a standard HTTP API that existing tools can call with minimal change.
- Hardware guidance. A written recommendation for the machines needed for your expected load, or a review of what you already own.
- Access and logging. Authentication, role limits and logs that stay on your side.
- Runbooks. Plain instructions for restarting, updating, backing up and rolling back, written for your operations staff.
How it is done
- Review the environment. Network layout, security rules, available hardware and how users will reach the service.
- Size and prepare. We choose the serving setup and, for sites with no internet connection, agree how files will be moved in.
- Deploy and connect. We install, integrate with your applications and run the same test set used during distillation, to confirm nothing changed.
- Hand over. We train your team on the runbooks and stay available for a support period we agree up front.
What we need from you
- Access to the target servers or a cloud account, granted on terms your security team sets.
- A named technical contact for network and security questions.
- A list of the systems that should call the model.
- For sites with no internet connection, an approved way to move files in.
Questions, answered.
Does it work with no internet connection?
That is the design goal: a model on your own servers needs no outside API to answer. We have not yet delivered an offline or air-gapped installation, so we would scope and test this with you first.
What hardware does it need?
It depends on the model and your load. Compact models often run on a single server with one GPU. We give you a written sizing recommendation after scoping.
Can our existing software use it?
Usually with little change. The service exposes a standard API of the kind most AI tooling already speaks, so switching is often a configuration edit.
Other services
- DistillationTrain a compact model for one task of yours.
- HostingA hosted AI desktop on hardware we operate.
- Agentic systemsAgents for multi-step work, under your approval rules.
Start on a practice company.
Request an invite and try it on made-up customers first. Nothing reaches real people while you learn.