Press Release: Locai Labs launches Locai One powered by NVIDIA Nemotron 3.5 Lightning

Locai Labs today launches Locai One, a compact on-prem AI appliance designed to bring advanced AI in-house using NVIDIA AI infrastructure and NVIDIA Nemotron open models. Locai One provides a practical and cost-effective route to own and customise enterprise AI by providing a private AI data centre in your office.

Locai One comes equipped with the new open-source Juno-N-Coder-25B model developed by Locai Labs.

Juno-N-Coder-25B is based on NVIDIA Nemotron 3.5 Lightning open model also released today.

Locai Labs compressed the 30-billion-parameter Nemotron 3.5 Lightning model to 25 billion parameters using its proprietary new SPACE algorithm (Specialisation Pruning Algorithm for Compression of Experts) so that it retains its competitive performance for coding and agentic use.

London, United Kingdom - 11 August 2026. Locai Labs Limited today launches Locai One: a compact on-prem AI appliance powered by NVIDIA, designed to enable organisations to move their AI workloads in-house, saving money and protecting their IP.

Locai One is an on-prem AI system built on NVIDIA AI infrastructure. Configured with NVIDIA Blackwell, it combines the new Locai OS (an AI Operating System), Locai Labs' new Juno models, an optimised inference engine and enterprise AI governance tools in a single appliance. It connects to an organisation's existing IT network and can operate in an office or server room, providing the AI capacity to serve the organisation’s existing AI applications and workloads.

“To own and control AI deployment and accelerate AI adoption in regulated industries, we need to make advanced AI models small and efficient enough to run where the data is,” said James Drayson, Chief Executive Officer of Locai Labs.”

“Compress the model enough and the data centre becomes the AI box sitting in the corner of your office. Locai Labs' compression algorithms make it possible to move your AI in-house so you can own and control it” said George Drayson, Chief AI Officer of Locai Labs.

Locai One provides a practical route to own and customise AI for enterprises. Instead of sending prompts, documents and agent workflows to an external cloud provider, organisations can run a private large language model within their own network, keeping sensitive data, intellectual property and operational knowledge under their complete control. And saving money by eliminating token-based charges.

Optimising Juno-N-Coder-25B on NVIDIA Nemotron 3.5 Lightning

Locai Labs' research focus is open-source model optimisation; improving the performance of advanced AI models, while reducing their size, to enable them to run on “the edge” locally.

The first Juno model launching today on Locai One is based onNVIDIA Nemotron 3.5 Lightning, an open mixture-of-experts model for reasoning, coding, planning and tool use. Locai Labs has reduced the model's total parameter count by 19%, from 31.56 to 25.57 billion, while retaining the base model's performance on coding and agentic tasks using its proprietary new domain specific expert compression algorithm (SPACE). The compressed model was then quantised, to NVFP4 for on-device inference on Blackwell GPUs.

Nemotron 3.5 Lightninguses a mixture-of-experts architecture, instead of running every parameter for every token, a router sends each token to a small handful of specialised sub-networks, so only a fraction of the model is active at any moment. Expert selection is skewed by input domain, but that skew is statistical rather than a clean task partition, so identifying what is safe to remove is the hard part. Locai Labs' proprietary SPACE algorithms score each expert by its contribution to a target domain, in this case software development, and prune those that contribute least. The result is a model with a much smaller memory footprint and the retained capability concentrated where we need it.

Current results for the Juno-N-Coder-25B model are:

• Pruned experts by 25%, achieving a total parameter reduction of 19%, from 31.56 to 25.57 billion parameters.

• Base-model performance retained on LiveCodeBench v6, HumanEval+, MBPP+, and SciCode after pruning with degradation in the areas outside its intended use including general knowledge and factual recall.

Juno-N-Coder-25B is the first of a range of new specialist frontier models under development by Locai Labs that are designed to provide the speed and performance required for demanding specialist workloads when run on the Locai One. Versions for use in scientific research, finance, law and healthcare are planned for later this year.

Why in-house AI is now needed

As AI becomes embedded across businesses and public services, organisations are considering whether every prompt, document and workflow should be processed through infrastructure controlled by an external provider in the cloud.

The reasons are practical. Running AI in-house allows organisations to keep sensitive data and intellectual property inside their own network, integrate AI models more closely with internal systems and retain control over how the technology is governed. It also provides a route to AI that you can own, and govern at organisational scale: the models, data and infrastructure on which an organisation depends remain under its control.

The economics are better too. Cloud AI services typically charge according to usage, so costs rise as more employees, workflows and AI agents are added. Locai One runs with a predictable, fixed hardware and software cost and no token-based charges.

Trials Across Government and Industry

Locai One is currently being trialled by the UK Government’s Incubator for AI to assess how sovereign, on-premises AI can support secure, data-sensitive workloads within government-controlled infrastructure.

"The Incubator for AI is pleased to be trialling Locai One and to support the kind of homegrown AI innovation that companies like Locai Labs represent, it's a great example of Britain's AI sector at work" said Max Hollingdale, Head of Applied AI Engineering, Incubator for AI, UK Government

As of today, Locai Lab’s UK sovereign AI assistant GB1 (www.gb1.ai) runs on Jupiter-N-120B model built on Nemotron 3 Super by Locai Labs and hosted locally on-prem from a Locai One NVIDIA GPU machine in the Locai Labs offices in London, further reducing its carbon impact.

About Locai Labs

Locai Labs is a British AI lab building sovereign AI models and systems for on-premises and edge deployment. Its research focuses on model compression, high-speed local inference and continual learning. Its flagship system, Locai One, is an AI Appliance to allow enterprises to bring AI in house.