Investing in Cercle

The team

Three co-founders, one senior advisor from a major IT player. One shared vision.

• President

  Co-founder, Paris

  An international career with solid software grounding. Product and commercial execution: scoping offers, running deals, client relationships. Ties to Station F and to the industrial fabric of the Paris region.

• Chief technology officer

  Co-founder, Nagoya

  Ten years in defence and aerospace with France’s major prime contractors. Specialist in industrialising high-performance embedded systems. Experienced in European public procurement, international technical cooperation and critical supply chains. Based in Nagoya, Japan.

• Head of business development

  Co-founder, Paris

  Arts et Métiers engineer, graduate of Sciences Po and of IAE Aix-Marseille. A defence background, experienced in strategy, business development and digital project delivery.

• Senior advisor

  Adviser

  Former head of Microsoft Defence, twenty-seven years in defence IT.

Our thesis

Generation is becoming a commodity, trust is not.

Models already produce text, search for information, synthesise a corpus and chain actions together. Improving that single capability further solves only part of the problem.

In a company, a recommendation is worth something only if someone can act on it. The more consequential the decision, the more that person must be able to understand what information it rests on, check its provenance, spot contradictions, weigh what remains uncertain and explain afterwards why a decision was taken.

When that verification takes as much work as the original production, the gain AI promised disappears.

That is the problem we address. We are working to scale verification rather than generation alone.

Our design principle follows directly: the human remains responsible for the decision. Beyond is not meant to hide that responsibility behind an autonomous system. It organises the preparatory work needed for the human decision to be taken quickly, explicitly and on defensible grounds.

This approach strikes us as particularly relevant in environments where an error engages a clear responsibility: critical industry, defence, energy, regulated finance, compliance, risk management, business intelligence or supply chain surveillance.

Product and value proposition

Beyond builds the case. The human decides.

Beyond is an agentic framework designed to produce what we call a Human Review Package.

A Human Review Package is not simply a model answer with a few sources attached. It is a case file built for a given class of decision. It gathers what human arbitration requires: the proposed result, the supporting evidence, the provenance and freshness of the information, any contradictions, the level of confidence, the useful context and what remains to be examined.

In an industrial surveillance case, for example, the system can detect a signal concerning a critical supplier and hand the professional the identified event, the items that support it, their freshness, the contradictory information and the relevant courses of action. The analyst no longer has to reconstruct the whole file before being able to take a position.

Our value proposition rests on that difference: we do not sell more information to verify. We work to reduce the cost of verification itself.

An architecture in four parts

1. 01

   The Enterprise Knowledge Backbone

   The EKB is the memory of the system. It treats knowledge not as a set of definitively established facts, but as beliefs that are sourced, dated, versioned and revisable. A piece of information can change, be contradicted or lose its validity without its history disappearing.

2. 02

   The Core

   It is the production and control layer. It handles identities, rights, data access, observability and traceability. It also organises case-building according to the standards of the trade concerned.

3. 03

   The Companions

   These are the business agents that work on top of this infrastructure. They inherit strictly the rights of the person they assist: the agent never becomes a way around the company’s access policies.

4. 04

   The Human Review Package

   This is the object handed to the user for arbitration. It is the interface between the work of the machine and human responsibility.

The whole is designed to be deployable in the cloud, in a sovereign environment or on premises, and not to depend structurally on a single model vendor. The choice of language model remains a parameter of the architecture rather than its foundation.

Research, patents and intellectual property

We base our product on internal research, and we are not afraid to put it before the judgement of our peers. Our approach is rigorous and scientific, and we draw a clear line between what is a scientific hypothesis, what has been measured experimentally and what is demonstrated in production.

The papers presented below are an extract from our ongoing research on the structural components of GenAI agents: we do not merely use commercial models, we build our own architectures in order to control every performance parameter. Our products are built on hybridising our in-house technologies with commercial models, so as to improve the precision, the fidelity and the exhaustiveness of what our models return. We are also working on the technologies underlying World Models, in order to exploit some of their properties applied to LLMs.

Three patents are also drafted and ready to file, on agentic memories and GenAI execution harnesses.

The company’s model

We are not trying to sell seats. We sell a case already built.

The economic unit we want to sell is not access to an interface. It is an operated business Companion, able to produce Human Review Packages continuously for a given business process.

The Beyond platform is the means of production of that service and a large part of our technological barrier. It is not, in itself, the product we ask the client to administer.

The model has three levels

1. 01

   Set-up, or build

   It covers scoping the domain, connecting the necessary sources and formalising the case-building standards. It is invoiced up front.

2. 02

   The Beyond platform

   Made up of the EKB and the Core in particular, it maintains the memory, the access rights and the explainability mechanisms. It is a recurring component.

3. 03

   Operations, or run

   They cover the continuous work of the Companions and their improvement. They are recurring as well.

We keep the option of billing partly on usage or on outcome where the outcome is objectively attributable and we control the inputs sufficiently. We refuse, however, to make the economics of the company depend on a promise of results that external data or a human decision could put beyond our control. The recurring platform therefore remains the base of the model.

Where we stand

We are at pre-seed stage. Here is what is established today, with no favourable rounding.

1. 01

   The company operates from two countries

   France and Japan, with the time coverage that gives and the commercial presence it requires on both sides.

2. 02

   Three letters of intent signed for a pilot

   Three companies have signed a letter of intent for a pilot project of our solution. A letter of intent is neither a contract, nor a promise to contract, nor a purchase commitment, and it carries no exclusivity: it says a need is recognised and a budget is being discussed. The amounts below are indicative, and the signatories stay anonymous, confidentiality being the counterpart of their signature.

3. 03

   The internal reference is written and applied

   It is versioned, it provides for its own annual audit, and every one of its rules can be invoked by anyone, including against the founders.

The round

We are raising €500k in pre-seed equity.

The round funds eighteen months of runway, the time it takes to turn our design partners into measurable proof.

• 500 k€

  in equity

  The pre-seed round itself.

• 100 k€

  non-dilutive

  Public schemes targeted: BFT, JEI status, research tax credit.

• 18 months

  of runway

  The time to turn our design partners into measurable proof.

• 370 k€

  Team

  Four full-time equivalents, two in Paris and two in Nagoya, founders included. Half the team in Japan, where the loaded cost is markedly lower.

• 50 k€

  Infrastructure

  Compute, hosting and production tooling.

• 40 k€

  Structure

  Company costs, including the creation and upkeep of the Japanese entity.

• 40 k€

  Go to market

  Pre-sales, travel and qualification of target accounts.

The trajectory

1. 01

   Seeding, from T0 to T0+6

   Closing the pre-seed, JEI status, BFT and research tax credit filings. First engineers hired, two to three design partners engaged.

2. 02

   Proof of the model, from T0+6 to T0+18

   Converting pilots into recurring revenue and signing contracts with contracted expansion tiers. Five to ten clients.

3. 03

   Scale, from T0+18 to T0+36

   Stepping up the pace and extending to the other markets identified. The seed round is then prepared on recurring revenue and the schemes that come with it.

The €100k of non-dilutive funding is not secured: these are schemes the company is eligible for, whose filings go in during the first six months. The plan does not count them as revenue, and the runway holds without them.