On-premise AI
Private LLMs and local models, on your own hardware.
We deploy language models on-premise, in your private cloud or fully air-gapped, so sensitive data never goes to a public AI API. Hospitals, universities, banks, government departments and any organisation whose data must stay inside: we size, tune, integrate and support the models, with no per-query fees.
Deployment options
Wherever your data has to stay.
The same models and tooling run in each setting.
On-premise
On servers in your own data centre.
Private cloud
Inside your own cloud account or virtual private cloud, under your security controls.
Air-gapped
On networks with no internet connection, for the most sensitive work.
Edge and CPU
Small task-specific models that run on ordinary computers, with no GPU.
What we deploy
More than a model on a server.
A useful private AI system needs the model, the knowledge around it and the controls on top.
Language models
Open-weight LLMs, or domain-adaptive models trained on your documents with Jungroo AI Studio.
Retrieval with citations
Search over your documents and systems, so answers point to their sources.
Agents and integrations
Connections to email, ticketing, document stores and core systems, with fixed limits on what agents can do.
Access control
Role-based access, aligned with your existing identity and security setup.
Audit logs and monitoring
Every request and answer recorded inside your network, with alerts on anything unusual.
Updates and support
Regular corpus updates and new releases, each validated before it goes live.
Hardware and cost
Smaller than you might think.
A model adapted to your domain does not need to be the largest model available. That keeps hardware modest and running costs predictable.
One GPU server
A 30-billion-parameter-class model in FP8 runs on a single 80 GB GPU server.
CPU-only models
Our ReDefiNER, a 0.8-billion-parameter model, extracts information from text on an ordinary CPU.
No per-query fees
You pay for the hardware, not for every prompt. Costs do not rise with each extra question.
Questions
On-premise AI, explained.
Can we run an LLM on-premise?
Yes. Open-weight language models can run on your own servers, in your private cloud or fully air-gapped. A model adapted to your domain is often smaller, so it needs less hardware than people expect.
What hardware does a private LLM need?
It depends on model size and the number of users. As a reference, a 30-billion-parameter-class model in FP8 runs on a single 80 GB GPU server, and small task-specific models such as our open-source ReDefiNER run on an ordinary CPU.
Is a local model as accurate as a cloud model?
For a well-defined domain task, a model adapted to that domain can compete with large cloud models; for open-ended general questions, the largest cloud models usually lead. We test this on your own data with a free two-week benchmark.
How is an on-premise model kept up to date?
Through regular corpus updates and new model releases, each validated before deployment, as part of ongoing support.
Contact
Deploy with precision.
We partner with a limited number of enterprises at a time, so every engagement gets senior attention. Brief our leadership on your roadmap.