Exploring how AI actually works
A UK lab built by someone who operates under real accountability. We test frontier AI systems, build with them, and publish what we learn.
Local Infrastructure
NVIDIA Blackwell, DGX systems and multi-GPU configurations. Running frontier models on our own hardware.
Agentic Workflows
Building AI agents for real work. Claude Code, automation systems, multi-step pipelines.
Open Research
We publish what we learn. Observations, experiments, things that surprised us.
Focus
Understanding AI from the inside
We run local AI infrastructure, build agentic workflows, and work with frontier systems daily. Not benchmarks. Real use. We are interested in what happens when you rely on these systems for consequential work.
Our research is informed by doing: deploying systems, observing behaviour, documenting what we find.
What We Do
Local AI Infrastructure
ActiveRunning frontier models on local hardware. NVIDIA Blackwell, DGX systems, multi-GPU configurations. Complete control over what we test and how.
Agentic Workflows
ActiveBuilding and testing AI agents for real work. Understanding where they succeed, where they fail, and what robust design looks like.
Reliability Research
ActiveDocumenting how AI systems behave in practice. Failure modes, edge cases, the gap between claimed and observed capability.
Recent
2026-01-26
What Breaks When Agents Run
Observations from building agentic workflows. Where they fail and what we have learned.
Global AI Activity
AI systems are being deployed. Right now.
We are building with these systems, testing them, and documenting what we learn. Research grounded in practice.
Work With Us
Interested in what we are building?
We are open to conversations about research collaboration, advisory work, and support for what we are building. If you find this interesting, reach out.