Private AI built around your business.

Identifying the subject of an incoming request is different from assessing complex technical documents. A compact model may be enough for one task, while another needs more capability. We select a model for the problem, rather than defaulting to the largest or most expensive option.

Larger models generally need more memory and computing power. When you host them yourself, that often means more powerful, more expensive servers. The extra cost needs to earn its place through better results or capabilities a smaller model cannot adequately provide.

If a smaller model does the job well, we choose it. More demanding tasks within the same application can use a more capable model. That keeps the more expensive capacity focused on the work that needs it.

We compare models using questions and documents from your business. Does the model understand the terminology? Does it recognise missing information? How much correction does an employee still need to make? We assess the results with the people who will use them, including difficult cases.

We also estimate costs at your expected usage. How often will people use the application, how many will use it at once and how quickly do they need an answer? We account for hosting, usage and ongoing support. A low model price means little if employees spend their time correcting results or waiting.

Useful AI often needs access to product information, customer agreements or internal knowledge. We prefer open models that we can run ourselves, in a private environment for your business. That is what we mean by private AI. You decide which information is available and who can use it.

A compact model may run well on a virtual private server, or VPS. A more demanding application may need more capacity. Self-hosting is not automatically cheaper: a server costs money even when it sees little use. That is why we choose the model and hosting together.

We favour European models where they meet the needs of the task. We also choose where your information is processed and who can access it, including the hosting provider. A European model does not automatically run in Europe.

An external AI service remains an option if it is a better fit. We make clear which information it receives, on what terms and at what cost.

Calculating a price or forwarding an approved order can follow fixed rules in software. AI can summarise a document while leaving the employee to decide what to do with it. Neither needs an agent.

An agent chooses its next step as it works through a task. For an unusual customer request, it might look up existing agreements, identify missing information and then retrieve further details. That can help when the sequence cannot be defined in advance.

We use this approach when the necessary information is available and those extra steps genuinely help the employee. We limit what the agent can do, make errors visible and recoverable, and agree when a person takes over. The benefit must justify the additional cost and ongoing support. The aim is to give people more ability to act, not to make the software as independent as possible.

Preparing a proposal is different from committing to a price or changing an order. We enforce those permissions in the software, not just in instructions to the model. Employees need to see what a proposal is based on and approve it where required. The same security and access controls apply to AI.

We bring experience of translating business agreements into software. For an application and partner platform, we built calculations and permissions around commercial terms. In JobLog, we connected customer agreements, recurring work and invoicing. We build on that experience when adding AI.

You receive a clear recommendation on what AI should do, which model fits and what the application will cost to run and maintain. This can be a standalone consulting engagement, including when your own team will handle the build.

If we also handle implementation, we build the application, connect the information it needs and set up hosting and ongoing support. Before release, we test with your employees to check it works as intended. We then track quality, usage and cost so it continues to serve the work.