Glossary

Sovereign AI glossary

The key terms behind sovereign AI, defined in plain English.

An AI model running on infrastructure with no connection to external networks, so the most sensitive data can be processed without any possibility of it leaving. Read more

CLOUD Act

A US law allowing American authorities to compel US-based providers to disclose data they control, even when that data is stored outside the US, including in the UK. Read more

Continual learning

An approach where a model keeps learning from new data over time without forgetting what it already knew, so it compounds in value rather than going stale.

Data residency

Keeping data, and the inference that processes it, within a defined geographic or legal jurisdiction. Residency is about location; sovereignty is about control. Read more

Domain-specific LLM (DSLM)

A large language model post-trained on a particular organisation's or industry's data so it reasons like an expert in that field, often beating larger general models on its tasks. Read more

Fine-tuning

A light adaptation of an existing model to a narrow task or style. Quicker than deeper specialisation but shallower, and prone to catastrophic forgetting. Read more

Forget-Me-Not

Locai Labs' post-training framework that adds deep domain expertise to a base model while preserving its general capabilities, mitigating catastrophic forgetting.

Inference

The process of running a trained model to produce an output for a given input (a prompt). With sovereign AI, inference happens inside your perimeter.

Mixture of Experts (MoE)

A model architecture that routes each input to a subset of specialised sub-networks ("experts"), giving large capacity while activating only part of the model per query.

Model weights

The learned parameters that define a model's behaviour, the actual asset. Owning the weights means owning the model, rather than renting access via an API key. Read more

On-premise (on-prem) AI

Running AI models inside your own data centre or hardware, rather than calling a model over the internet, so data stays in and cost is fixed. Read more

Post-training

Deeply specialising a strong base model on your domain, language, and workflows while preserving its general reasoning, the approach Locai uses to build owned models. Read more

Private AI

AI that runs inside your perimeter with no external data sharing or telemetry, so prompts, documents, and outputs never leave your environment. Read more

Retrieval-augmented generation (RAG)

A technique that fetches relevant documents at query time and feeds them to a model. Complementary to post-training, but it doesn't change how the model itself reasons.

Sovereign AI

Enterprise AI you fully own and control, the weights, the infrastructure, the data path, and the update cadence, deployed inside your own perimeter rather than rented through an API. Read more

Sovereignty washing

Marketing a service as "sovereign" because it runs in a local data centre, while control still sits with a foreign provider, superficial sovereignty without real control. Read more

Book a sovereign AI briefing

A 30-minute session on owning your model: deployment options, the data path, and a clear cost range for your use case.

Book a sovereign AI briefing

Order a Locai One