FINPAR
Innovation Project · 10 sites live

3 independent projects

Three investment plans. Each project will have its own holding company, to be set up in Luxembourg, its own funding and its own exit. All are driven by artificial intelligence, which sells, answers, invoices and follows up day and night, weekends and public holidays included: the customer who buys at three in the morning is served just like the one at noon.

An investor may choose one, two or all three. In each case, the money is released in three conditional tranches and goes mainly to marketing: the services are built; what they lack is customers.

The investor holds 49% of the holding company, acquired as the tranches are paid: about 29% of the issued capital after the first, 43% after the second, 49% on completion of the third; the project is sold at the close of the sixth financial year. On the sale, the investor is first repaid 51% of their stake, the share the founders were unable to contribute in cash; the balance is then shared in proportion to shareholdings, 51/49. The returns shown ignore this priority: they are floors.

Our market studies give each of these projects real potential, but the ambition is not to dominate these markets: it is to take a modest, measurable share of them, by concentrating the investment on one thing: making services that already exist known, quickly.

The tests conducted to date have been carried out by the founding team; the beta-testing phase with real users opens at launch, driven by the first advertising budget.

The partnership between founders and investors is a one-off, and that is its strength: it seeks neither to build a group nor to last. It serves to launch each project, to specialise its artificial intelligence in its trade, and to bring it to a sale within five years. Everyone knows on the way in why they are there, and how they get out.

A specialised, profitable, self-running service sells well. That is what this partnership builds: at the sale, each project will be a proven business, with a measured customer base, clear accounts, and operations that depend on no single person. The plans assume a prudent valuation of one and a half times revenue; a specialised, demonstrated asset may command more, and no figure on this site assumes that premium: if it comes, it is additional.

Three projects, three exits, one method: measure, prove, sell.

Aductori: Sophia, a supportive presence assisted by artificial intelligence Discover Sophia
Iirmaa: Irma, clairvoyance, astrology and numerology guided by AI Discover Irma (site in French only)
Astrioni: one account, a whole ecosystem for the self-employed Discover Astrioni and its 7 sites
3
independent projects, three separate structures
€3.1 M
requested in total, spread across the three
€41.5 M
combined value of the three sales in the 6th financial year, if all three plans are achieved
49%
of each holding company held by its investor

What this site is, and what it is not

It is the entry point to three funding requests presented in a single proposal. It is not a fund, a collective vehicle or a pooled investment: each project has its own company, its own accounts and its own exit, and an investor enters one of them, not all three by default.

Why now?

General-purpose models are progressing fast, and some predict a major leap within two to three years. We do not treat this as a forecast; we draw three consequences from it.

A different economy is taking hold, in which the service is produced without labour: one more customer does not require one more employee, and the margins in these plans, from 70 to 78%, are not a promise but the cost structure these three services already have.

The cost of rented artificial intelligence is falling at constant quality, and none of these plans relies on that decline.

Finally, the window is closing. Establishing a specialised service, with its trade data and its customer base, becomes harder as general-purpose models improve: what can be defended in 2027 will no longer be defensible in 2030. This is what drives the timetable; it is not a sales pitch.

Who does the work, if the team is 3 people?

It is the question the salary scale raises: for example, the €10,985,600 of revenue that Astrioni's plan projects for the 6th financial year, handled by a handful of people, looks at first like a spreadsheet error. The explanation lies in the positions that are not filled, because the machine fills them. In an online services company, what can be delegated today covers most of the functions that, yesterday, determined headcount.

Development and technical operations. Writing code, deployment, server monitoring, fixing routine incidents. No developer position is budgeted in these plans, and no external development services appear in the forecasts.

Producing the service sold. This is the core: the reply to the customer, the consultation, the analysis, the document delivered. The service is produced on demand, within seconds, with no queue and no schedule to keep.

Advertising and editorial content. Visuals, ads, videos, pages, listings and sales copy. The advertising budget pays for media space, never for making the ads: there is no designer, no agency and no copywriter.

Languages. The 6 languages of the target market are handled without translators or local offices, for the product as well as for advertising. This is what makes a European market accessible to a team of 3 people.

Customer relations. Requests are handled 24 hours a day, every day, in the customer's language. Human intervention remains possible; it is the exception, not the normal way of working.

Administration and management reporting. Payment collection, credits, invoices, reminders, but also the monitoring dashboards: customer acquisition cost by cohort, usage, alerts. These are the figures on which the tranches are conditional, and they are produced automatically.

What remains for the team is of a different order: designing the service, building this machine, monitoring it, deciding on spending and handling what the machine declines to handle.

This is work that does not grow with the number of customers. The forecasts remain prudent on this point: the payroll budget funds up to 6 people in the 6th financial year, whereas the base case assumes 3. Even at 6, the broadest assumption, it leaves more than €1.8 M of revenue per person.

This organisation is not a projection awaiting funding. It is already the one that produced the three services, live today, before a single euro of investment and without a single payslip.

01 · The three projects

The three projects, side by side

They are listed in order of readiness, from the most advanced to the least advanced, not by amount.

Swipe the table to the left to see all three projects.

ADUCTORI
IIRMAA
ASTRIONI
Type of market Consumer Consumer Businesses
What is sold Round-the-clock listening, from €0.05 to €0.09 per question Divination consultations, from €3.50 to €15 Seven pay-per-use AI services, from €1 excl. VAT, minimum top-up €10
Product status Live, able to take payments now Live, 13 services delivered, could take payments now 8 sites live, 3 engines out of 7 ready for operation
Market European, 6 languages French-speaking for 12 months, then English-speaking for 12 months, then the rest European, 6 languages
Amount requested €900,000 €1,000,000 €1,200,000
Revenue at the end of the plan
(calculation method in the full business plan)
€9,776,600 €6,914,700 (French-speaking market only) €10,985,600
Exit value · 1.5 × revenue, low estimate €14,664,900 €10,372,000 €16,478,400
Investor's share · 49% of the exit value, before priority repayment of 51% of the stake (floors) €7,185,800 €5,082,300 €8,074,400
Return on investment 8.0× 5.1× on the French-speaking market alone 6.7×

A caveat for the reader

The Aductori and Astrioni columns are costed on six languages, Iirmaa's on one only. Opening the other five multiplies its addressable market by four, without a single euro of this being included in either its plan or its multiple.

The business plan, on request

A summary business plan covering the three projects is available on request: an overall presentation, then one plan per project, with the figures year by year, the fund release mechanism, the escrow, the team and the exit. Written in French.

Write to us through the Contact us form, the last link in the navigation menu, stating which of the three projects you are considering. We send it with the supporting documents within two working days.

See for yourself, now, without us

Best opened on a computer: that is where the sites are shown in full, animations included.

  • Aductori, aductori.com: the site is functional and its payment system is active.
  • Iirmaa, iirmaa.fr: the site, in French only, is functional and currently offers a beta-test version to its visitors.
  • Astrioni, astrioni.com and its seven satellites: all eight sites are live as showcase sites, and three business engines can be used directly.

Why would anyone pay for Sophia, when AI assistants answer for free?

This is an investor's first objection, and it is a legitimate one. The answer lies in the subject itself. What people confide in Sophia, they confide neither to someone close nor to anyone who knows them: they fear being judged, or that it will get out, or that it will one day be used against them. Until now, the other recourse was a psychologist: around €60 a session, and an appointment that can take weeks to get. Yet the need to talk does not wait. It arises at 11 pm on a Sunday. Sophia answers at that very moment, 24 hours a day, 7 days a week, for €0.05 to €0.09 per question. She does not present herself as a therapist and refers people to a professional when needed. She occupies the space between silence and the consulting room.

Compared with free services, the difference is not the price but the use. A general-purpose assistant changes stance from one question to the next and is bound by no framework. Sophia has only one role. She keeps the same stance throughout the exchange, remembers what she has been told and never makes a diagnosis. As soon as a message suggests danger, a safety net independent of the AI provides the country's emergency numbers. This continuity brings the user back, and it is on that return, not on the first question, that the credit model rests.

Then there is trust, which matters more the more one confides. The major assistants store the conversations of millions of people, most often on another continent, under other laws, and their terms of use can change. Aductori asks only for an email address and builds no advertising profile. It stores conversations in Europe, under the GDPR, and deletes them with the account. Drafting the replies still relies on external providers, under standard contractual clauses. The in-house model planned after the investment will end this, and will make confidentiality an argument that free services cannot match.

Astrology or tarot by AI: don't customers prefer a real clairvoyant, face to face?

Some do, and they will continue to consult one. But Iirmaa does not sell clairvoyance in the usual sense: no supposed gift, no promised prediction, no fictitious practitioner. Astrology, numerology and tarot reading are systems of symbolic interpretation, codified frameworks for reading passed down over centuries: the planets and their influences, the arcana and their meanings, the numbers and their vibrations. A consultation applies these correspondences to a person. These frameworks can be learnt and reproduced, and that is precisely what AI handles with consistency and depth. Each reading starts from the real birth data, calculated by our own engine, whereas the free horoscope serves the same text to everyone.

Face to face comes at a price: from €60 to €350 per consultation with a practitioner, by appointment, and €1.80 to €9.50 per minute on telephone platforms. Irma answers at any hour, 24 hours a day, 7 days a week, without an appointment, for €3.50 to €15 per full consultation. The plan does not count on the customers who will keep their clairvoyant: it targets all those for whom that price was a barrier. The industry estimates this market at €3 to €4 billion a year for France alone, with about 100,000 practitioners, and one adult in four says they have already consulted one (IFOP, 2020). The plan deliberately assumes a much lower base: €1 to €2 billion for the whole French-speaking market.

The ambition is not first place. A modest share of this market is enough for the plan to hold, and the investment goes mainly to marketing, since the service is already built. The timetable is the same as for the other two projects: investment, operation, sale at the close of the sixth financial year. The buyer then takes over a business with clear accounts and an AI specialised in its trade by six years of consultations. On the French-speaking market alone, the plan projects a return of 5.1 times the stake, before the other five languages are even opened.

Eight sites and seven services: isn't that too scattered?

The question is a fair one, and the answer lies in the architecture. Astrioni does not sell seven separate products: it sells a single account. The hub, astrioni.com, handles registration, credits and invoicing. Each of the seven satellite sites is an entry point, focused on a specific problem of a self-employed professional or a small business:

  • mapping one's local competitors and one's position in the local rankings (Arkhoni);
  • having one's website, Google reviews and brand presence on the web audited (Athenori);
  • having one's Google Business profile and social media posts written (Asteroni);
  • identifying what is missing around a location before starting or taking over a business (Actaoni);
  • drafting the right follow-up to a prospect, by email, SMS, WhatsApp, LinkedIn, letter or call script (Aedironi);
  • monitoring one's online reputation at the chosen frequency and replying to reviews (Argyroni);
  • building a contact base through a prize competition (Alearoni).

This is what drives the project's economics. A customer comes in through one door and discovers the other six: same account, same credits, no new registration. Each service they add increases revenue per customer without any additional customer acquisition cost. It is the only place in the model where an additional sale costs nothing. Seven doors also means seven ways of being found, each site having its own name and search ranking, where a single platform would have only one. There is no subscription, so nothing to cancel. Credits do not expire, and every eleventh identical service is free.

The scattering would be real if seven teams were needed. Not a single extra one is: the eight sites share the same technical base, the same account and the same invoicing. The AI performs in a few seconds what used to cost hours of consultancy or €100 to €500 a month in specialised tools, from €1 excl. VAT per use. All eight sites are live, and three engines out of seven can already be operated. At the sale, it is precisely this architecture that a buyer pays for: a portfolio of services that sell one another. The plan assumes no premium for this and projects 6.7 times the stake. Even with revenue per customer cut by a quarter, it still yields 5 times the stake.

High-risk investment

These are companies in the process of formation. The capital may be lost in its entirety. The figures presented are projections built on stated assumptions, not results. Nothing on this site constitutes a public offer, a solicitation or investment advice: the investment pack is provided on request to sophisticated investors, and any commitment is governed by a written contract.

02 · In detail

The three projects

Each is presented in order of readiness. For each: what it sells, where it stands today, what remains to be built, and what it is asking for.

ADUCTORI

An artificial intelligence that listens around the clock, for at most €0.09 per written exchange.

Sophia answers those who have no one to talk to at three in the morning: emotional dependency, hypersensitivity, identity, grief. She never claims to replace a doctor or a therapist, and says so.

Open aductori.com →
  • Live, and able to take payments now. The text chat is complete; Sophia is in open beta, each exchange can be rated by the member and this feedback is used to tune the model; audio and video remain to be developed.
  • Six languages, eleven countries covered, as of today. Opening up across Europe requires no development: the service is already translated and live.
  • €900,000 requested, released in three tranches of €390,000, €310,000 and €200,000.
  • Profitable from the 3rd financial year, first profitable month: month 20; peak cash requirement €579,400.
  • €9,776,600 of revenue in the 6th financial year, a 75% margin on net revenue (excl. VAT), and a customer acquisition cost of €12.
  • €14,664,900 valuation at the sale, i.e. 1.5 times revenue. This is the low assumption: the plan uses the lower of two methods, 1.5 times revenue or 6 times profit.
  • €7,185,800 for the investor, i.e. 49% of this valuation and 8.0 times their stake. The sale is initiated at the close of the 6th financial year only if revenue reaches at least €4,000,000; below that, it is postponed by two years and the project distributes half of its profit in the meantime.

IIRMAA

Divination consultations produced by artificial intelligence, delivered 7 days a week, from €3.50 to €15 instead of €60.

Astrology, tarot, numerology, birth chart: thirteen services already delivered, on a platform that is running and whose payment system is active.

Open iirmaa.fr (in French) →
  • Live, in beta testing, payment active. The service is open to the public and ready to leave beta.
  • The astronomical calculation is done in-house. The sky chart engine was written internally: no royalties and no obligation to open the code, unlike the standard industry library, whose licence requires it.
  • French only for twelve full months, then English only for twelve months, then the last four languages together. This is not a translation delay; it is a decision about support workload.
  • €1,000,000 requested, in tranches of €430,000, €350,000 and €220,000.
  • €6,914,700 of revenue in the 6th financial year, a 70% margin on net revenue (excl. VAT), on the French-speaking market alone. Not a single euro from the other languages is included in the plan.
  • €10,372,000 valuation at the sale, i.e. 1.5 times revenue. This is the low assumption: the plan uses the lower of two methods, 1.5 times revenue or 6 times profit.
  • €5,082,300 for the investor, i.e. 49% of this valuation and 5.1 times their stake. The sale is initiated at the close of the 6th financial year only if revenue reaches at least €4,100,000; below that, it is postponed by two years and the project distributes half of its profit in the meantime.

ASTRIONI

Seven artificial intelligence services sold per use to the self-employed, micro-businesses and SMEs, from €1 excl. VAT per use, minimum top-up €10.

No subscription, no quote, no salesperson: the customer buys an analysis, receives it, and leaves. It is the only project in this proposal aimed at businesses.

Open astrioni.com →
  • Eight sites live: Astrioni is the hub, where the account, credits and invoicing are kept, and gives access to the other seven. Arkhoni shows who your competitors are, Athenori analyses your website, Asteroni writes your posts, Actaoni tells you what is missing, Aedironi writes your follow-up messages, Argyroni monitors your reviews, Alearoni runs a prize competition.
  • Three engines out of seven can be operated directly; four remain to be finished, along with centralised payment and the administration console.
  • €1,200,000 requested, in tranches of €520,000, €420,000 and €260,000.
  • €10,985,600 of revenue in the 6th financial year, a 78% margin on net revenue (excl. VAT), the highest of the three, because business customers spend more per purchase.
  • €16,478,400 valuation at the sale, i.e. 1.5 times revenue. This is the low assumption: the plan uses the lower of two methods, 1.5 times revenue or 6 times profit.
  • €8,074,400 for the investor, i.e. 49% of this valuation and 6.7 times their stake. The sale is initiated at the close of the 6th financial year only if revenue reaches at least €5,000,000; below that, it is postponed by two years and the project distributes half of its profit in the meantime.

Each stands on its own without the other two

Each project will have its own company, its own accounts and its own exit. An investor may choose one, two or all three. Each budgets its complete set-up without assuming that another will be funded at the same time: this proposal may interest only one investor for only one project, and that project must then bear its incorporation costs and its lease on its own.

These amounts are projections, not results

These are companies in the process of formation, and the capital may be lost in its entirety. The sale amounts shown above assume that the plan is achieved and that the revenue threshold is reached; if it is not, the sale is postponed by two years and the project distributes half of its profit in the meantime. Nothing on this site constitutes a public offer, a solicitation or investment advice.

03 · The common mechanism

What the three projects share

Different markets, but the same mechanics underneath: how the money is released, how the service is invoiced, where the companies are established, and how an investor exits.

Swipe the table to the left to see all three projects.

Tranche When ADUCTORI IIRMAA ASTRIONI What it is for
1 Months 1 to 9 €390,000 €430,000 €520,000 Set up the companies, open the markets, measure customer acquisition cost
2 Months 10 to 18 €310,000 €350,000 €420,000 Strengthen the markets that respond, reach cruising speed
3 Months 19 to 27 €200,000 €220,000 €260,000 Secure cash flow, strengthen the team

€68,000 per project is held by a lawyer, not by us

The first tranche funds expenses incurred before any company exists: a notary cannot be paid from the account of an entity the notary has not yet incorporated. This is the window in which the investor is most exposed, and it calls for a dedicated arrangement: an escrow account held by the investor's lawyer, their usual lawyer, appointed before any tranche is paid. The incorporation invoices that follow (notary, accountant, tax lawyer, lease) are addressed to that lawyer, who pays them against supporting documents. None of these sums passes through us.

  • Tax lawyer, €30,000: structuring, intellectual property regime, shareholders' agreement.
  • Notary, accountant and disbursements, €20,000: incorporation, articles of association, capital deposit, chart of accounts.
  • Office rental, €18,000: twelve months prepaid at €1,500; the lease is a condition of the business permit.

These are ceilings, not quotes. Published hourly rates for the Luxembourg bar range from €300 for a junior associate to €545 for a senior associate: €30,000 therefore buys 55 to 100 hours. We prefer a generous ceiling to a call for further funds in the middle of the set-up, since whatever is not used returns to the company.

The balance is released against four proofs

This is not the investor's second tranche; it is the rest of the first. As soon as the founder establishes, with supporting documents, that the structure exists and is operating, the balance is paid into the operating company's account.

  1. 1. The companies are incorporated: extract from the Luxembourg Trade and Companies Register (RCS), articles of association filed, capital paid up.
  2. 2. The business permit is obtained. This is a requirement specific to Luxembourg: without it, no activity can legally begin, and it requires premises and a qualified manager.
  3. 3. The lease is signed and paid for twelve months.
  4. 4. A bank account is opened in the company's name alone, and is operational.

Two points are settled in writing before any funds enter the escrow, and they are settled with the investor: the name of the lawyer, who is the investor's own (their usual lawyer: an escrow chosen by the founder alone would be worthless to the person depositing their money) and a deadline, the date by which the four proofs must be assembled. After that date, the lawyer returns to the investor whatever they still hold, less only the sums already paid out against invoices. No automatic extension.

Summary

None of the three tranches is guaranteed in advance. Each is subject to a written condition, and an unmet condition stops the spending rather than continuing it.

  • The first is split into two parts. Only €68,000 goes into the escrow of a lawyer appointed by the investor. All the rest of the tranche, which is by far the larger part, reaches the operating company's account only once the structure has been established with supporting documents: companies incorporated, business permit obtained, lease signed and paid, bank account opened. A condition precedent is added: the sites, the domain names, the payment account and the contracts are transferred to the Luxembourg company that is to hold them.
  • The second is conditional on the customer acquisition cost actually observed over months 4 to 9, on real cohorts, and the threshold is specific to each project: less than €13 for Aductori, with a projected lifetime of at least twelve months; less than €39 for Iirmaa; less than €53 for Astrioni. This is the most important provision in the proposal: it places verification before the heavy spending, not after. Above the threshold, the project is renegotiated rather than continued, and advertising stops the following month.
  • The third secures cash flow and strengthens the team, at the point when the plan goes through its last trough. No numerical threshold applies to it, and we say so rather than invent one: it is the investor's decision, and the investor keeps control to the end. A tranche that is not released also freezes the salary scale, which rises only with the tranches.

What the investor holds, and what they can block

The 51/49 split applies only to the capital of the holding company; the holding company owns 100% of the operating company, which collects payments and invoices. It therefore amounts to 49% of the whole group, and no asset is held in a third-party entity.

49% is not a passive stake. In a Luxembourg public limited company (société anonyme), extraordinary decisions require a two-thirds majority: the investor therefore holds a genuine blocking minority, not merely a right to dividends.

A detailed term sheet is signed before any tranche is paid, and the shareholders' agreement is signed at the closing of the first tranche, at the same time as the first disbursement. The actual sequence is as follows: the term sheet sets out in writing the reserved matters, the timetable for the 49%, the exit and the distribution priority; the investor then deposits the €68,000 with their own lawyer, the one they already work with. We get in touch with that lawyer and send them the incorporation invoices, including the fees of the Luxembourg tax lawyer, and it is the lawyer who pays them, against supporting documents. This payment triggers the legal work, which produces the articles of association, the contribution deed and the shareholders' agreement, signed at the closing of the first tranche.

What this means for the investor, plainly: if they stop after this first stage, the sums already paid out by their lawyer against invoices are not returned to them. These are costs actually incurred, not a provision. It is the only point in the arrangement where they commit without a guarantee of return, and it concerns €68,000, not the whole tranche: everything that has not been paid out is returned to them by their lawyer on the deadline, with no automatic extension.

What artificial intelligence requires of us

The three services rely on generative models sold to individuals and businesses in the Union: they fall within the scope of the EU AI Act, whose obligations have been phasing in since 2025.

Three product choices follow from this, and they predate the regulation. The use of artificial intelligence is disclosed, never hidden behind a fictitious practitioner. Iirmaa is positioned as personalised entertainment, not as prediction. Sophia makes no diagnosis, recommends no treatment, and tells the user so: the limit is enforced in the model itself, not only in the terms of use.

The same approach to artificial intelligence

All the services currently call external models through an API. Two separate providers are permanently connected and genuinely share the requests, with automatic failover from one to the other in the event of an outage, a missing key or a timeout. Neither is a dormant backup whose working state is unknown: an outage or a price rise at one of them stops no service.

An in-house model is planned after the investment. The aim is not raw performance, but turning a variable cost into an asset: rented artificial intelligence accounts for 14 to 22% of revenue and grows in step with revenue. The work is carried by the first employee, whose position is already in the salary scale, and is conducted on the existing infrastructure: it opens no line of expenditure that the forecasts do not already carry. No effect of this project is included in the forecasts, which bear the external cost over the whole period.

This transition is not an architectural gamble. Conversations are stored in our databases, as text, with no vector index that would need rebuilding, and the gateway sends no conversation identifier to the provider: the context is rebuilt with each call. Aductori and Iirmaa already switch provider mid-conversation, in production, with the same memory. What does not transfer is not the data but the voice: the same framing text on another model does not produce exactly the same character, and this recalibration is planned for.

Credits bought in advance, never a subscription

Customers buy what they use and stop when they wish. No recurring direct debit, no cancellation to remember, no forgotten charge. This costs us a steadier revenue line, and in return buys us longevity: no one feels trapped.

Luxembourg, and what it costs

One holding company per project, corporate income tax at the aggregate rate of 23.87%, partial exemption of up to 80% on intellectual property income. Substance is not declared, it is paid for: real premises, a real lease, a manager on site. It costs more than a low-tax jurisdiction, and it withstands scrutiny.

A small team, paid below market

€6,000 gross per month for a founder, when the Luxembourg market for these profiles ranges from €7,000 to €12,000. A tranche that is not released freezes the salary scale.

One employee per company at launch, so two per project, one in the holding company and one in the operating company: two of the three founders, including the qualified manager on whom the business permit depends. The third is hired at the second tranche, earlier if the workload requires it. The salary scale is a ceiling, not a commitment: €6,000 gross per month per founder at launch, then at most €6,500 and €7,000 as the team grows, i.e. no more than €94,500 of employer cost per position, employer social contributions included. Any salary paid below it stays in the company. This is the ceiling budgeted for the founders, and the budget is not limited to them: it funds up to six people in the sixth financial year, with hires made after the three founders budgeted at no more than €4,000 gross per month, so that reinforcement does not require going back to the investor. The base case assumes three. The difference feeds the advertising budget and is nowhere counted as a gain. Experienced artificial intelligence profiles command €13,000 to €15,000: the salary scale adopted is below the market's entry level, not merely at its lower end.

Before becoming an employee, the third founder provides support without pay. This is not a cash-flow device: it is their contribution, the one they cannot make in cash, and it is this same imbalance that the priority repayment at the sale corrects.

The published salary scale

This is the "salaries and expenses" line of the simulator, broken down year by year. It applies per project, each having two companies: the holding company and the operating company.

Financial yearHeadcountGross salariesSocial contributionsExpensesTotal
1st (3 months)2€36,000€4,500€27,400€67,900
2nd3€166,500€20,800€64,600€251,900
3rd3€247,500€30,900€71,200€349,600
4th4€300,000€37,500€105,800€443,300
5th5€348,000€43,500€147,500€539,000
6th6€396,000€49,500€164,800€610,300

The founders' salaries follow the tranche rule: €6,000 gross per month until the second, €6,500 thereafter, €7,000 from the third. Luxembourg employer social contributions are counted at 12.5%. Hires made after the three founders are budgeted at no more than €4,000 gross per month: this is what allows headcount to grow without the payroll running away.

Expenses in the first financial year include the initial equipment: servers and switches for €18,000 and workstations for €7,000. Rent is not included: the first twelve months are prepaid through the escrow, so it appears only from the thirteenth. The following financial years include the rent, the fiduciary firm and the accounts of both companies, the statutory auditor, hosting, tools and insurance.

This salary scale is a budget, not a commitment to hire. The team grows as needed, and the base case costs less than the budget. The difference feeds the advertising budget and is nowhere counted as a gain.

The exit, at the close of the sixth financial year

The investor holds 49% of the holding company once the three tranches have been paid. The valuation uses the lower of two methods (1.5 times revenue and 6 times profit), and revenue is the determining factor in all three projects. If the revenue threshold is not reached, the sale is postponed by two years and the project distributes half of its profit in the meantime.

04 · About

Who is behind this?

What exists today, and what we do not claim.

The team, and why it is not named here

Two of us are still bound by a permanent employment contract. As long as these contracts run, this site gives no names, no career histories and no employers.

This is not a reservation on substance. CVs, references and each person's role are presented in person at the first meeting, then documented. It is an ordinary precaution, and it follows the same rule as the rest of the proposal: nothing irreversible is committed before the structure is.

Years of trials before the first euro requested

There are three of us, from the same family, all web developers, and we have followed artificial intelligence since it left the laboratories. We did not study it by reading: we studied it by producing, because a tool only shows where it gives way when it is asked to deliver.

Eight virtual authors, each with their own genre and catalogue, and some sixty books published under their names (bookorama.eu). Around a hundred music videos (projets-ia.org), including an educational series for children (c-pafo.com). These three sites are in French only. None of this work was designed to make money. All the books were indeed put on sale on Amazon KDP, but for a reason other than revenue: to obtain the judgement of readers who had paid, the only kind no internal trial can provide. They were test benches, and each one identified a specific limit.

The music videos served for a test that text alone did not allow: that of sung lyrics. A few bilingual songs, others tested in seven different languages. A wrong pronunciation can be heard, whereas a clumsy translation is barely noticed in reading.

This is not done in a few clicks. It is a complete chain, and at the time it ran without agents, directly with Claude or ChatGPT: translation of the lyrics by artificial intelligence, then composition and performance by a music generation program, then creation of the still images, then their animation, then the video editing. Five different trades, five different tools, and one weak link is enough to spoil the whole. It is by linking these steps together that one learns where each gives way, and it is this experience that produced the method described below.

The most decisive thing is not what we produced, but what we observed from one series to the next: the limits noted the previous time had disappeared. We saw artificial intelligence progress on our own work, month after month, from one version to the next. It is this acceleration, more than any demonstration, that made up our minds.

Then came the artificial intelligences specialised in software development. We have since worked with Codex and Claude Code, among others, and we drew from this the conclusion that governs everything else: an artificial intelligence is now capable of developing and running an entire business on its own, provided it is led by people who know how to tame it, or rather to prompt it (the prompt is the written instruction it is given: everything else depends on its precision), and who can read what it writes.

The three projects presented here are the test of this conclusion. They are neither mock-ups nor intentions: they are built, they are live, and they are open to investment.

What we lack is neither technical skill nor the ability to run these businesses: it is capital. They were built and put online through our own work alone, and it is to carry them through to financial self-sufficiency that we are approaching investors.

What an agent is, and why we gave each one a trade

Agents came later, and they changed everything. An agent is not a conversation with an artificial intelligence. It is a trade assigned to it once and for all: a written mandate, rules, an output format, and a prohibition on doing anything else. It is not asked for an opinion on the whole; it is entrusted with one task and relieved of all others.

Each works alone, without seeing the work of the others, and that is precisely what makes its verdict usable: it has not been influenced by the one before it. A proofreader who had written the text would not find their own mistakes.

When we launched them, we trained them on books. The production of the last titles published on Kindle Direct Publishing, Amazon's self-publishing platform, served as their school: one title after another, the same agents recalled as needed and corrected frequently. An agent that has held up over dozens of books is a tuned agent.

There are eight of them, they pass the manuscript along in this order, and what comes out at the end of the chain is a complete book, typeset and ready to publish:

  1. The editor leads the work: framing, editorial plan, decisions, approval of the move from one version to the next, and the metadata for the Kindle listing.
  2. The researcher prepares the fact sheets, sources and chronologies, and hands them over before the first line is written.
  3. The writer writes the chapters in the style of the series, under explicit rules against the writing tics peculiar to artificial intelligence.
  4. The first reader reads critically and scores on six criteria: hook, pace, clarity, interest, emotion and naturalness. At the same time, they track down repetitions.
  5. The fact-checker checks names, dates, places and chronology, verifies the facts and cross-checks the chapters against one another.
  6. The art director sets the structure of the book, the box-outs, the visual specifications and the cover options.
  7. The proofreader makes the final pass, accents, punctuation, grammar, sentence by sentence.
  8. The final checker steps in after the deliverable file is generated: this is the last barrier, and any residual error sends the file back for regeneration.

Then each site received its own agents, created on demand, for what the book chain could not judge:

  • Translator agents, one per language, for the six languages of the sites. Translating is not the difficulty; the difficulty is translating the same thing six times without the six versions diverging.
  • Anonymous customer agents, which visit a site as an ordinary visitor would, try to buy, and report in detail on what they understood, what they could not find and the exact point where they gave up.
  • A professional clairvoyant agent, which judges Iirmaa's services on the only ground that matters to those who consult them: how a consultation holds up, from the point of view of someone for whom this is their trade.
  • And others besides, which we do not detail here. The skill does not lie in the tool, which anyone can buy, but in writing the mandate: it is the only asset of the team that cannot be picked up by visiting the sites. It is shown at the first meeting.

This team is not paid, does not wear out, and is called back for the next project. It is this team, not a bet on technology, that explains how three people were able to build ten sites, nine of them in six languages.

The technical foundation is written here, and it is public

One of us wrote the framework on which the ten sites run. A framework is the foundation of a site: what manages the pages, accounts, payments, languages and security, beneath what the visitor sees. This one is built for applications where security comes first, with an audit log, performance budgets and a modular architecture.

It is published free of charge under an open-source licence on GitHub, the public platform where developers publish their code and where anyone can read it, check it and reuse it. Nothing is hidden: the repository address is provided with the investment pack, and the code can be audited by your own technical adviser.

This choice has three direct consequences for an investor. The foundation is not rented from a third-party vendor, so no supplier can cut off access to it or raise its price. A security fix written once protects all ten sites on the same day, instead of being redone ten times. And code exposed to everyone's view is code that other eyes review, which no team of three people could otherwise afford.

It does not belong to the projects, and this must be stated clearly. The framework remains the property of the person who wrote it; the three companies will use it free of charge. This does not make it a dependency: the open-source licence under which it is published is perpetual and irrevocable. It cannot be withdrawn, charged for later or refused to anyone, the three companies included. No royalty is due today and none can be in the future.

Nor does it earn anything, and this is deliberate: as long as the projects generate no revenue, its distribution is worth more than a price. No revenue from the framework is included in the plans of the three projects, which expect nothing from it.

What carries the sites today

Putting the sites online and opening a payment account required an existing legal entity. For this we used an English company that one of the founders still has at their disposal, FIRST AGENCY LTD, dormant for years and kept in good standing. It therefore has no past operations to take over and no ongoing commitments: it serves as a placeholder shell, nothing more.

It is a transitional vehicle, and we say so plainly: the three projects will each be transferred to their own Luxembourg holding company, and the first tranche funds precisely these incorporations.

Why there is no marketing yet

It is not a choice; it is a consequence of the lack of funding. A thousand euros of advertising produces nothing: the sum is diluted before it has allowed anything to be measured, and spending that cannot be measured serves no purpose.

That leaves organic search, that is, what raises a site in search results without paying anything. It is free, but slow: it takes six to twelve months before a site brings a steady flow of new customers, and eighteen to twenty-four months when it is new or the market is contested. Ours are new.

The only lever that shortens this delay is paid marketing, and that is exactly what this proposal is asking for.

Without advertising, these sites are needles in the haystack that is the web. They work, they can take payments, and almost nobody knows they exist. This is not a product defect: it is precisely what the money buys here. The platforms are built; what remains is to make them known.

What we do not claim

  • That everything is finished. Four of Astrioni's seven engines remain to be finished, along with centralised payment and the administration console. Aductori's audio and video are still to be developed.
  • That the figures are results. They are projections built on assumptions written into the plans, and the assumption on which everything depends (the cost of winning a customer) is measured before the heavy spending, not assumed.
  • That the market will reward us for existing. None of the three plans assumes taking a significant share of its market. What they assume is that the customer acquisition cost measured in the ninth month holds at cruising speed, and that is where the risk lies.

A full business plan exists for each project

The documents provided are summary versions: they state the essentials and can be read without any particular financial expertise. The projects are listed in order of readiness, from the most advanced to the least advanced.

If a project catches your interest, a full investment pack is provided on request: the monthly financial model, the detailed sensitivity analysis, the competitive landscape, the capitalisation table, the tax regime and the list of points still to be validated. The calculation model itself can be opened and re-run in front of you.

Understanding artificial intelligence before judging the projects

A guide in its 2026 edition, explaining what artificial intelligence is, what it can already do and where it stops. It is not a sales document: it is there so that an investor can form their own judgement on what underpins these three projects. The guide is currently available in French only.

Discover the AI guide (PDF in French, 169 pages) →
05 · Contact

Contact us

The investment pack is provided on request. Tell us which of the three projects you are considering: we reply within two working days and attach the supporting documents.

Links are not accepted in this form: please describe your request in full. Your contact details are used to reply to this request, and for nothing else.