Est. 2024Paris · France

Building AI systems for real-world work.

LucidQuery is a sovereign European AI lab, based in France. Since 2024 we've shipped two generations of hybrid models, a desktop command layer, and developer tools. Our flagship is AGI 01 — a model family driven by the World Engine, the orchestration harness we built in 2025.

Try LucidQueryExplore the API
Chronology

Two model generations, then the engine.

Every model we shipped taught us the same thing, and AGI 01 is what we built once we believed it. The lit part of the track is the work still ahead.

  1. 2024Origin

    Founded in France

    A founder out of one of France's grandes écoles assembles a small team of AI researchers around one question: where does intelligence actually come from?

  2. 2024Models

    LucidNova RF1-100B

    Our first hybrid: diffusion in the reasoning space, autoregressive in the response engine. A 100B general-purpose model, and proof the two architectures were better together.

  3. 2025Models

    Nexus Coder

    A coding specialist built the other way round — diffusion-based generation driven by autoregressive reasoning, tuned end to end for software work.

  4. 2025Platform

    The World Engine

    Two model generations taught us the leverage wasn't in the weights. The World Engine is the orchestration layer that wraps around any model and multiplies what it can do, no retraining required.

  5. 2026Flagship

    AGI 01

    The model family everything else was building toward: Swift and Frontier, driven by the World Engine, behind one OpenAI-compatible API.

  6. NextFrontier

    AGI you can run at home

    The harness already delivers a tenfold lift. The next bet is size: post-training plus the World Engine so small, self-hostable models punch far above their weight — frontier-class work on hardware you already own.

Architecture

AGI 01, and the engine underneath it.

AGI 01 is what you actually use — Swift and Frontier, behind one API. What makes it strong is what sits under it: the World Engine, the orchestration harness we built in 2025 after two generations of hybrid models.

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.

LucidNova RF1-100B and Nexus Coder came first, and they are why the harness exists — two hybrid architectures that showed the leverage was never in the weights alone.

  1. Model-agnostic core
    route to any engine
  2. Orchestration layer
    plan, decompose, delegate
  3. Tools, memory & live web
    wired in, not bolted on
  4. Amplified intelligence
    markedly smarter output
Products

A lab's worth of AI.

The World Engine leads the way, but it sits on years of models, tools, and research shipped from our lab in France.

01

AGI 01

The model family: Swift for everyday speed, Frontier for the hard problems, both scored on our LucidQuotient benchmark and served behind one OpenAI-compatible API.

02

The World Engine

The orchestration layer underneath AGI 01. It wraps around any model and multiplies its intelligence, no retraining required.

03

Nexus Coder

Coding specialist: diffusion-based generation driven by autoregressive reasoning, tuned end to end for software work.

04

LucidNova RF1-100B

Our first hybrid — diffusion in the reasoning space, autoregressive in the response engine. The research the whole stack is built on.

05

LucidArc

A command layer for Windows: search, compute, and call AGI 01 one keystroke away.

06

LucidCode

An agentic coding CLI that brings AGI 01 to the terminal — plan, edit, and run across a whole codebase.

07

StrikeLoc

Pinpoints where a photo was taken through a ranked, auditable chain of visual evidence.

Principles

Principles that shape the work.

01

Amplify any model

Wrap the harness around a model and its capability multiplies, most on the hard, multi-step problems where it counts.

02

Model-agnostic

Open weights or frontier APIs. Swap the engine underneath at will.

03

Ultra-connected

Tools, memory, live web, and your data, orchestrated as one layer.

04

Transparent & European

Built to European standards, sovereign by design. Every step the harness takes is legible and auditable.

Foundation

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.

2024
Founded in France
5+
AI products shipped
Hybrid
Diffusion-AR research roots
European
Independent & sovereign
Experience it

See how AI should really work

Try the hybrid architecture for yourself. Ask it something hard and watch it reason.

Start a conversationBuild with the API
Talk to our team