Building AI systems for real-world work.
LucidQuery is a France-based AI lab. Since 2024 we've shipped models, a desktop command layer, and developer tools. In 2026, our flagship: an ultra high-end harness that wraps around any model and multiplies what it can do.
The harness: one layer, around any model
We stopped betting on a single architecture. Instead we built a harness, an ultra-connected orchestration layer that wraps around any AI model and multiplies what it can do.
It routes between engines, wires in tools, live web access and persistent memory, and drives multi-step reasoning, turning a capable model into a markedly smarter system, without retraining a thing.
Our roots are in hybrid diffusion-autoregressive research; the harness is where that work pays off.
A lab's worth of AI
The harness leads the way, but it sits on years of models, tools, and research shipped from our lab in France.
The Harness
Our orchestration layer wraps around any model and multiplies its intelligence, no retraining required.
AGI-01 & AGI-02
Frontier reasoning models, Swift to Frontier, scored on our LucidQuotient benchmark.
LucidArc
A command layer for Windows: search, compute, and call AGI 01 one keystroke away.
Nexus Coder
A specialised model for generation, debugging, and review across languages.
StrikeLoc
Pinpoints where a photo was taken through a ranked, auditable chain of visual evidence.
From a French lab to a global platform
A short history, in orbit. Tap any milestone to expand it.
Principles that shape the work
Amplify any model
Wrap the harness around a model and its capability multiplies, most on the hard, multi-step problems where it counts.
Model-agnostic
Open weights or frontier APIs. Swap the engine underneath at will.
Ultra-connected
Tools, memory, live web, and your data, orchestrated as one layer.
Transparent & European
Built to European standards. Every step the harness takes is legible and auditable.
A small team, a long bet
After graduating from one of France's prestigious grandes écoles, our founder assembled a team of researchers and engineers around one idea: the biggest gains wouldn't come from a single model, but from the layer wrapped around it.
We pair deep technical work with a product sensibility, turning research into tools developers actually ship. The result is an ultra-connected harness that makes any model measurably smarter.
Operating from France, we lean on European standards of responsibility and transparency: technology that advances what people can do, without asking them to take it on faith.
See how AI should really work
Try the hybrid architecture for yourself. Ask it something hard and watch it reason.