How can self-hosted AI give organizations more control over data?

When an AI model processes sensitive information, organizations need to know where that processing happens. They also need control over model changes and how the service operates.

For Geotab, the motivation is specific. Our platform processes telematics data, trips, driver behavior, location, vehicle health, on behalf of fleet operators who treat that as competitive intelligence. 

Self-hosting ensures that data processed by the model stays inside our perimeter. It also means GIA, Geotab’s internal AI assistant, runs on infrastructure we own — not a commercial API that can be rate-limited, updated without notice, or taken offline. For a platform serving 24/7 fleet operations, that control is not a nice-to-have.

Our pilot explores three forms of control:

• Model control: Choose when to change model versions and behaviour.
• Data control: Keep sensitive prompts within the stated perimeter.
• Service control: Reduce reliance on a hosted model API that may change or become unavailable.

These controls can matter to fleet operators because their data describes vehicles, routes, driver behaviour, and business operations. Knowing where that information is processed can help organizations assess whether an AI service fits their privacy and operating requirements.

Does self-hosting make an AI service safer and more stable?

Self-hosting can provide more control over the model and data path. It does not automatically make a service safer or guarantee that it will remain available.

The organization must still manage access, monitoring, maintenance, and the infrastructure that serves the model. Self-hosting changes where some dependencies sit. It does not remove every supplier or service risk.

For Geotab, the pilot is one way to explore more control over AI operations. It is not a claim that the pilot cannot fail or that it replaces every hosted model.

What does the serving platform contribute?

The model is only one part of an AI service. The surrounding platform must route requests, detect failures, and show whether the model is completing useful work.

One lesson from our pilot is that a service can appear healthy while routing requests incorrectly. That can undermine predictable service behaviour and resource use. A useful health check should confirm that the model completes a real request, not only that a process is running.

These are the practical lessons for fleet and business leaders: ask how the system handles model changes, where data is processed, and how teams know the service is working. The detailed routing and monitoring design is for technical readers.

What does this pilot mean for customers?

Geotab’s self-hosted LLM work is an experimental pilot in Geotab’s commercial environment only. It is not authorized for FedRAMP or other regulated government workloads. It is not an announcement of a customer-facing self-hosted model.

The pilot reflects a broader commitment to control over how AI processes sensitive information. Customers should continue to assess the data boundaries, security controls, and service commitments of any AI capability they use.

For the technical builder’s account of routing, cache behaviour, and monitoring, read The Model Is the Easy Part: What We Learned From Self-Hosting Large Language Models Open in new window.

Related reading: The AI Tools Are Ready. Is Your Data?