LINAGORA AI
Our positioning

Our model: produce, integrate, operate managed inference

We do not sell access to a technology we keep to ourselves. We produce digital commons, we help you put them to work, and we operate them for you for as long as you wish.

The three layers of the LINAGORA model, producing the commons, advising and integrating, operating and hosting, each with what you buy and what you are left with.

What a digital common is, and why it changes everything

A digital common is not free-of-charge software. It is a resource produced and maintained collectively, whose evolution its users can govern. Being free of charge is sometimes a consequence, never the definition.

Applied to artificial intelligence, the term commits to four very concrete things: downloadable weights, code that can be audited line by line, documentation of the training chain that says where the data came from, and a community whose existence does not depend on the survival of a single company.

That last condition is the least cited and the most decisive. Software published under a free licence by a vendor that disappears remains legally free and becomes technically dead. A common stays alive because several organisations have an interest in maintaining it. That is why we do not merely publish code: we cofound and we run the collectives that carry it.

The commons we produce

Four building blocks, four distinct purposes, all published and maintained in the open.

Luciole

Language models trained in the open, at 1, 8 and 23 billion parameters, weights published under Apache 2.0.

Maturity : Version 1.1 publicly available, together with the base models and the quantised builds.

Open-RAG.ai

Querying document corpora: ingestion, vectorisation, retrieval and sourced answers.

Maturity : In production in public and private organisations.

LinTO.ai

Speech processing: transcription, translation, meeting summaries, in real time or after the fact.

Maturity : A mature platform, in production for several years.

Twake AI

Sovereign collaboration suite augmented by AI: messaging, video meetings, documents, business agents.

Maturity : In production, notably within European institutions.

The collectives that carry them

We cofounded OpenLLM France, the open community behind the LUCIE and then Luciole models. We organise the Paris Open Source AI Summit, which brings together the European players of open AI. We work with NVIDIA on the Nemotron family to train open European models. These commitments are not communication: they are the condition for the building blocks we publish to outlive you, and to outlive us.

OpenLLM Europe lists more than eighty open language model projects, across some thirty European countries. View the OpenLLM Europe catalogue

The three layers, in detail

Each layer stands on its own. Taken together, they produce a property the sum of the parts does not have: reversibility.

01

We produce the commons

Luciole, Open-RAG, LinTO, Twake AI. These building blocks are not products we rent to you. We publish them, we maintain them, and anyone can take them back, audit them or evolve them without asking our permission.

What you buy
Nothing. The building blocks are public, downloadable and usable without going through us.
What you are left with
Software and weights that nobody can take away from you, whatever becomes of our commercial relationship.
02

We advise and we integrate

Upstream strategic advice, technical expertise, integration. We help you decide before you deploy, then we adapt these building blocks to your context, your data and your regulatory constraints. Twenty-five years of free software running in demanding production environments.

What you buy
Engineering and consulting time, and documented deliverables.
What you are left with
A system adapted to your data, and the documentation that lets a third party take it over without us.
03

We operate managed inference

Bespoke managed AI services, hosted on platforms subject neither to the Cloud Act nor to any other extraterritorial law. You do not simply take a building block, you take an operated service, with a service level commitment and genuine contractual reversibility.

What you buy
A service operated to your level of requirement, with a service level commitment.
What you are left with
Your data, your weights and your configuration, exportable in their original format.

Why few players hold all three

The AI market splits into three trades, and most players practise only one of them.

Matrix of four kinds of player compared across three trades. The vendor, the integrator and the host hold one or two. LINAGORA holds all three.
The vendor

It operates and it integrates, but it produces no commons. Its technology belongs to it. You depend on its roadmap, on its pricing policy and, ultimately, on its survival.

The integrator

It advises and it deploys, but it does not produce the building blocks. It assembles what others own, and it inherits the limits of what it assembles without being able to lift them.

The sovereign host

It operates, but it neither produces nor deeply understands what it hosts. Where the machines sit settles a legal question, it does not settle a technical one.

LINAGORA holds all three positions: model builder, integrator, operator of managed inference. This is not a sales argument, it is what makes your exit possible: an organisation working with us can decide, at any time, to take the building blocks back and operate them itself, because they are open. That is the whole difference between sovereignty you own and sovereignty you rent.

What this guarantees you

Reversibility is not a contract clause, it is a property of the arrangement. It requires three conditions at once, and two out of three will not do.

  1. 01

    Open building blocks

    The code and the weights are public. You can download them, audit them and run them elsewhere. Without this condition, reversibility comes down to the right to leave empty-handed.

  2. 02

    Hosting outside extraterritorial law

    Processing, backups and logs all sit on platforms subject neither to the Cloud Act nor to any other extraterritorial law. Without this condition, your data remains reachable by a jurisdiction you did not choose.

  3. 03

    Documented integration

    The adaptations built for you are delivered with their code, their documentation and their operating procedures. Without this condition, the system is open but unreadable, and you cannot hand it to anyone else.

What these commitments rest on

25 years
publishing and integrating free software in production
200 engineers
in Paris, Tunis, Montreal, Brussels and Hanoi
European institutions
production deployments at the European Commission and the European Parliament
French ministries
production deployments across several ministries

Two ways to start a conversation

Depending on whether you still have to decide, or already know what you want to run.