Private AI: local models on your own infrastructure, so the data stays inside your network.
When data must not leave your network, we run the model where the data is: open-weight models such as Qwen, Llama and Mistral on your own servers, on-premises or in your own cloud account. Prompts, documents and answers stay inside your network. We size and build the infrastructure the model runs on, connect it to your systems the same way as a public model, and operate it afterwards: updates, monitoring and capacity.
When a local model is the right choice
- Regulation rules out an external API. Some data may not be processed outside your own infrastructure.
- Contracts do. Customer or supplier agreements forbid passing their data to a third party.
- Internal policy does. Designs, formulas, source code or financial data that the company has decided never to send out.
- If none of these apply, a hosted model is usually the faster start: see Claude and OpenAI integration. Many companies use both.
What we build and run
Model
An open-weight model that fits the task: Qwen, Llama, Mistral and others.
Infrastructure
We size and build the servers the model runs on, on-premises or in your own cloud account.
Serving
The model is served inside your network; prompts, documents and answers do not leave it.
Connections
The local model is connected to your ERP, documents and databases the same way as a public model, behind your SSO.
Monitoring and capacity
We watch load and response times and plan capacity as usage grows.
Updates
Models and the serving stack move on; we keep both current.
Hosted or local: the difference
| Hosted model | Local model | |
|---|---|---|
| Models | Anthropic Claude, OpenAI GPT, Google Gemini | Qwen, Llama, Mistral and other open-weight models |
| Where prompts are processed | By a cloud provider, under contract | Inside your network |
| Infrastructure | None of your own | Your own servers, on-premises or in your cloud account |
| Who operates it | The provider runs the model; we run the integration | We run the model, the infrastructure and the integration |
Customer project
- Presimsan: for Presimsan, a manufacturer, we built local AI this way (reference)
Frequently asked questions
What is private AI?
A language model that runs on infrastructure you control, on-premises or in your own cloud account, instead of at an external provider. Prompts, documents and answers stay inside your network.
Which local models do you run?
Open-weight models such as Qwen, Llama and Mistral, chosen for the task and for the infrastructure available.
Do we need our own servers?
The model runs on your own servers, on-premises or in your own cloud account. We size the infrastructure for the model and the expected load, and build it.
Is a local model as capable as Claude or GPT?
The most capable models today are hosted ones. Local models handle many company tasks well, such as questions over documents, drafting and classification. The choice is usually driven by where the data may go, which is why many companies use both.
Who operates the model after it is built?
We do: updates, monitoring and capacity, by the team that runs production clouds.