Locai One
The On-Prem
AI Computer.
Fixed-cost on-prem hardware that runs your models, right on your desk. Trained on your data, served to your whole team, with nothing ever leaving your environment.
Locai Model Customiser
Build your own
enterprise AI model.
Locai Model Foundry builds a bespoke AI model for your business. We post-train a frontier model on your industry and your proprietary data, then deploy it inside your environment, ready to serve your team through the applications we provide. You own the model, the weights, and the pipeline.
Train
Post-train a frontier-class base model on your proprietary data and workflows.
Deploy
Run on Locai One, your cloud, or our UK sovereign cloud, weights stay yours.
Own
The model, weights and pipeline are yours, no vendor lock-in, no per-token bills.
Customer Case Study, First Light Fusion
"We've been impressed by the quality and speed of the work delivered by Locai Labs."
Mark Thomas
CEO, First Light Fusion
First Light Fusion is an Oxford University spinout developing inertial confinement fusion energy. Locai Labs deployed FLF Coder, a domain-specific coding model, air-gapped and trained on their scientific and engineering workflows, to accelerate their fusion programme.
Research
Sovereign AI that
learns over time.
Our research is in continual learning in production, recursive, self-improving systems where a model can teach itself from real-world use. Jupiter is the frontier foundation we post-train and improve on your data through Locai Model Customiser.
Our Models
Our sovereign post-training of NVIDIA's Nemotron-3-Super-120B, adapted for British context with stronger instruction following, agentic behaviour, safety, and Welsh language capability, without regressing the base.
Hugging Face
The same Locai post-training recipe applied to Google's Gemma-4 base, lifting instruction following, safety and code while preserving general knowledge.
Our flagship British post-trained model. A 235B-parameter Mixture-of-Experts instruction model based on Qwen3, post-trained with Forget-Me-Not™ and aligned with UK laws, languages, and cultural context.
Our models
Frontier Performance.
Every Locai model starts at frontier-grade capability. Benchmarked, openly published, and engineered to retain the base model's strengths when we adapt it to your domain with your data.
Request API AccessJupiter-N
Oursvs. Nemotron base
All values in %. Reasoning on/off configurations match those reported in the Jupiter-N technical report. Welsh benchmarks evaluated reasoning-off.
Jupiter-G
Oursvs. Gemma-4 base
All values in %. Same Locai post-training recipe (Forget-Me-Not™ replay + curated SFT) applied to a different open base. General knowledge held within ~1.4 pts of base while gains stack on instruction following, safety and code.
Locai-L1-Large
Oursvs. Frontier average
Locai-L1-Large compared against the leading frontier models per benchmark. Sources: Arena Hard v2 (GPT-4.1 judge), IFEval and BritXNLI.
The benefits of
owning your AI.
Pay for the infrastructure once, not per prompt.
Fixed-cost deployment with no per-token bills. Run unlimited inference inside your environment, and the model gets sharper, and more valuable, the more your team uses it.
Your data, your IP.
The weights, the training data, and everything the model learns stays yours. You retain full ownership of the IP.
Your model and your data stay put.
The model runs inside your sovereign perimeter, in the jurisdiction you choose.
Trained on your business.
Post-trained on your documents, your workflows, and your domain, so it understands your context the way no off-the-shelf general model can.
Smarter every retrain.
Your model retrained using our Forget-Me-Not™ framework. The more you use it, the more it knows about your business, creating an asset that compounds in value.
Full audit trail.
You have full access to the weights, the training data, and the post-training process. With logs and traceability.
Built using British
science and know-how.
Locai Labs is a British company headquartered in London.
