Lab · Real systems, experiments in progress

A product lab building custom autonomous AI systems.

Software that does real work inside a business: it watches what's happening, decides, acts and checks the result. Below, six of them at work: some open to the public, some running inside our own operations.

Madorlive
Quota24live
Enzolive
Q24 Wireinternal
R.P. Systeminternal
CAÇAinternal

See the systems

How we build

Three rules we build by.

We ship, we don’t pitch

We don’t write strategy decks. We ship software that reads the data, decides and acts. What we release works on real data from the first day; what isn’t ready yet is labelled demo or concept.

Small surface, deep stack

A small interface on top of layered agent systems. You see only what you need to decide; the work happens underneath, on its own where it’s safe, with a person where it matters.

Built to operate on command

Our systems run on schedules: cron jobs, scheduled agents, retries when a step fails. They keep working at night and on weekends. You set the rules: what runs by itself, what waits for a person to approve it, what runs only when you ask.

Built · Running · Result

Six systems at work.

Three are public, three are internal. Each window is drawn from the real interface with example data, and says what is switched on and what isn't yet.

Mador

Live·public on mador.ai

Shows a local business whether AI assistants name it when a customer asks for what it sells, and who they name instead.

While running: It asks the assistants the questions a customer would ask, reads every answer, and marks who gets named and which sources the answer leaned on.
Where it stands: Public on mador.ai: anyone can run a free scan. The paid plans repeat the measurement over time.

Enter Mador

mador.ai · reportExample data

Costa Adeje · seafood restaurant

Fictional business

We asked the assistants what a customer would ask, and looked at who comes up.

Out of 10 ways of asking, you come up in 0

AssistantQuestionOutcomeProof

ChatGPTbest seafood restaurant in Costa Adejenames youpos. 23 sources
Perplexitybest seafood restaurant in Costa Adejesearched, didn't find youverifiable sources
ChatGPTwhere to eat fresh fish near the harbourdoesn't remember youanswer from memory
Perplexitywhere to eat fresh fish near the harbournames youpos. 45 sources
ChatGPTromantic dinner by the sea, Costa Adejebrushes past yousearched, sources not shown
Perplexityromantic dinner by the sea, Costa Adejenames youpos. 12 sources
ChatGPTseafood restaurant open on Sunday eveningdoesn't remember youanswer from memory
Perplexityseafood restaurant open on Sunday eveningsearched, didn't find youverifiable sources
ChatGPTfish restaurant with a sea viewsearched, didn't find yousearched, sources not shown
Perplexityfish restaurant with a sea viewnames youpos. 34 sources

The rest of the report

Who comes up instead of you, the sources behind each answer, what to do next.

Quota24 family

The product, the agents behind it, and an agent that plays on it. Same markets, three systems.

Quota24

Live·public on quota24.com

Our Italian prediction market. People take a side on what happens next in politics, the economy, sport and tech, and the price shows what the crowd expects.

While running: Every market has a question, a closing date and a price from 0 to 100 that moves as people take positions. When it closes, it's resolved with a stated source and positions are settled in Q Points.
Where it stands: Public on quota24.com, in Italian. It runs on Q Points, a play currency: no real money changes hands.

Open Quota24

quota24.comExample data

Search markets…

AllPoliticsEconomyFootballTechCrypto

Open markets

Will the ECB cut rates by December?

50%50%

YesNo

New31 Dec

Will gold top $5,400 an ounce by 31 December 2026?

43%57%

YesNo

1 Jan

Will Apple launch Siri AI in the EU by the end of 2026?

52%48%

YesNo

31 Dec

Q24 Wire

Internal·runs every day behind Quota24

The agent system behind Quota24. Agents read the news, propose markets, draft articles and check how markets ended. An editor approves before anything goes out.

While running: Every day Pulse proposes new markets from the news and Scoop drafts the articles. When a market has closed, Verdict searches for the outcome with sources and resolves it only above a confidence threshold.
Where it stands: It works every day behind quota24.com. Every market and every article is approved by a person before it's published.

quota24 · admin · Q24 WireExample data

3 new items to approve

PulseAutomatic market generationScoopAutomatic news generationVerdictAutomatic market resolution+ 7 agents

Pulse suggestions waiting: 0Generating…

Economy

Will the ECB cut rates by December?

Type: binary · Closes: 31 Dec

YesNo

“A decision is due in December and the topic is debated: the outcome isn't a given.”

ApproveReject

Verdict · outcome: Yes · confidence 91% · fact + source

Enzo

Live·paper account and Q Points

An agent experiment on Quota24's markets and on Polymarket. It forms its own estimate of a market and enters only when the price is far enough from it.

While running: It picks a market, reads independent sources, writes down its estimate and its reasoning, then enters or passes.
Where it stands: It plays with Q Points, Quota24's play currency, and on paper on Polymarket: the moves are real, the money isn't. Its log is public, losses included.

Watch Enzo's log

quota24.com/tomExample data

Newsroom agent

Enzo

Looks at a market, forms a view and sets it next to the price. If they're far apart it enters; if not, it passes and says why.

Gym, account in Q Points

Polymarket, paper account

PlaysPassesAll

Arena · paper accountenters
Will the ECB cut rates by December?
Estimate
61 out of 100
Price
46 out of 100
Gap
+15
The gap is wide and three independent sources point the same way.
What it read · 4 · where they agree and where they don't
Gym · Q Pointspasses
Will the derby end in a draw?
Estimate
31 out of 100
Price
28 out of 100
Gap
+3
Too close to the price to be worth a position.
What it read · 3 · where they agree and where they don't

R.P. System

Internal·sending switched off

The operations system behind our own short-term rentals in Tenerife. At its core is the Stay OS co-host: it reads each guest message, works out which apartment it's about and drafts the reply.

While running: Every thirty minutes it reads the Booking and Airbnb inboxes without touching them, matches each message to its apartment and writes a draft in the guest's language, using only what's known about that apartment.
Where it stands: Sending is off on purpose. Every reply is still a draft that a person reads and approves, and a draft overtaken by a newer message can't be sent.

R.P. System · InboxExample data

R.P. SystemOperations · short-term rentals

DRY-RUN · sends are simulated, nothing actually goes out

BookingENarrivingpending

Apartment in knowledge base

Documents
Registered
Code
Check-in

guest

Hi! Our flight lands at 23:40. Is a late check-in OK?

Host draft · reply to the latest message

Approve

Airbnbobsolete · a newer guest message arrived

CAÇA

Internal·private tool

A scouting tool for real estate. It cross-references public data sources and ranks the buildings worth a closer look.

While running: It rebuilds its list from public data and gives every building a score from 0 to 100, with badges you can read in a second: who runs it, how it's split, what the land allows.
Where it stands: We use it, we don't sell it. The dossiers it prints carry public data only, never the owners' personal details.

caça · explotadorExample data

ExplotadorTakeoverDistressedConstructionGestión

Public sources
Building found
Signals
Score
Why it's worth a look

Catastro
Registro Turístico
BORME
+ other public sources

Anonymised building · example

Tourist complex · 16–40 units

No single operatorDormant

78priority score

No single registered operatorRegistro Turístico
Tourist / hospitality use on recordCatastro
Linked companies checkedBORME
On tourist landplanning

Why it's worth a look

A tourist complex with no single registered operator, on tourist land: a candidate for single management. A person checks it before anything else.

Public data only

The operational layer

Behind every system we run, a team of agents.

Some watch, some decide, some act, some raise the alarm. Where a mistake would cost something, a person approves first.

IllustrationSimulated activity with sample data from five sectors. The four roles are how our real systems are split.

PULSE · INPUT AGENT

SIMULATED

HOSPITALITY·INPUTSIMULATED

New booking received · 2 guests
Inquiry: family of 4 · summer week
Check-in scheduled: 18:00 today
Booking modified: extra night added
Direct reservation completed
3 new reviews to acknowledge

VERDICT · ANALYSIS AGENT

SIMULATED

HOSPITALITY·ANALYSISSIMULATED

Classifying: "Optimal weekend pricing"

└─ Sources analyzed: 22
└─ Historical pattern: HIGH
└─ Confidence model: 84%

Result preview:

BUMP +12%

84%

HOLD

16%

Resolved · 11m ago

BUZZ · OUTPUT AGENT

SIMULATED

HOSPITALITY·OUTPUTSIMULATED

Drafting · Welcome message

Weekly digest #47

3 releases you missed
this week — full reviews
inside.

Industry trends, deep
dives, and one signal
worth watching.

Read more →

ALERT · REACTION AGENT

SIMULATED

HOSPITALITY·REACTIONSIMULATED

Cleaning crew needed: 11:00 deadline
Late check-out request received
Price gap detected vs competitor -8%
Review pending response: 3 days
Maintenance request: plumbing urgent

ACTIVITY · SIMULATED FEED

SIMULATED

14:32:18PULSEscanned 14 sources (PUBLISHING)SUCCESS
14:18:42VERDICTclassified pricing strategy (HOSPITALITY)SUCCESS
14:05:11BUZZdrafted client update (VETERINARY)PENDING
13:47:33ALERTflagged pricing gap (RENTAL OPS)WARNING
13:21:09LENSclassified 12 articles (PUBLISHING)SUCCESS
12:54:27TRENDdetected regulatory drift (PHARMA)WARNING

The pattern across our systems

Observe, understand, decide, act, verify, learn.

01
Observe
Mador asks AI assistants the questions a customer would ask, and records who gets named.
02
Understand
R.P. System works out which apartment a booking email is about before writing anything.
03
Decide
Enzo compares its own estimate with the market price, then enters or skips.
04
Act
R.P. System drafts the reply to the guest in the guest's language.
05
Verify
A person approves each message. Stale drafts can't be sent.
06
Learn
Mador measures again and compares with the previous reading.

It isn't a shared platform. It's the pattern our systems have in common, and the direction we're building in.

THE SYSTEM AT WORK

One flow, end to end. Illustrative traces with example data from five sectors: the shape of a run, not a production log.

▶ KHLUMEN · session #2847

PUBLISHING

22:04:12

session: 23s · steps: 0 · result: —

BEFORE / AFTER · EXAMPLE

What changes, concretely.

An illustrative diff of a check-in workflow. The numbers are an example, not a client result.

hospitality-ops/src/check-in/workflow.tsx

+3 · −3 · 10 lines

export const CheckInWorkflow = () => {

  return (

    <Pipeline>

−

      <HumanWelcomeMessage delay="4h" />

−

      <ManualInboxCheck channels={3} every="evening" />

−

      <PhoneTriage avgWait="22min" />

      <DraftReply lang="guest" time="<2min" />

      <HumanApproval required />

      <FreshnessGuard maxAge="2h" />

    </Pipeline>

);

};

Example<2minto draft a reply·1approval per message·reply in the guest's language·3inboxes, one queue

The Lab

Everything we've started, and how far each one got.

Some of it runs every day, some is a demo you can click through, some is still an idea. The label on each one says which.

In operation

Each one is shown in full in Systems.

Live
Mador
AI visibility for local businesses
Measures whether AI assistants name a business, and who they name instead.
See the system
Internal
R.P. System
Rental operations
Runs guest messaging for our own short-term rentals: the Stay OS co-host reads guest messages and drafts replies, a person approves them.
See the system
Internal
CAÇA
Real-estate scouting
Cross-references public data sources and ranks the buildings worth a closer look. Private tool.
See the system
Live
Quota24 family
Quota24
Prediction market
Italian prediction market, public. People take a side with play money; the price shows what the crowd expects.
See the system
Internal
Quota24 family
Q24 Wire
Agent operations · Quota24
The agents behind Quota24: they propose markets, draft news and check outcomes. An editor approves everything.
See the system
Live
Quota24 family
Enzo
Agent experiment · Quota24
An agent that forms its own estimate and plays only when the price is far from it. Paper account and Q Points, public log.
See the system

Demos

Interactive mockups with sample data, further down this page.

Demo
Stay OS
Hospitality OS
The hospitality panel we're designing from what runs in R.P. System: pricing, messages, turnovers.
Open the demo
Demo
Pitch
Football player market
Players traded like assets, prices moved by transfers and matches. Simulated market.
Open the demo
Demo
AI Compliance
EU AI Act compliance
A registry of a company's AI systems, with risk classes and an audit trail.
Open the demo

Concepts

Ideas we've worked on. None of them is a product yet.

Ink AI
Concept
Tattoo studio OS
The artist tattoos; agents take care of bookings, reference material and admin.
Pharma AI
Concept
Pharmaceutical operations
Document workflows for pharma teams, with every step traceable for audits.
Kayfabe
Concept
Wrestling fantasy platform
Storylines, prediction markets on rivalries and an AI coach, for wrestling fans in Europe.
Hitbound
Concept
Emerging music
A market on emerging artists, priced on streams and trend signals.
Wire Network
Concept
Multi-vertical editorial
Hyper-local newsletters, each vertical gathered by agents and edited by people.
Vet AI
Concept
Veterinary operations
Operational AI for veterinary practices, built around clinical workflows.
IT Skill Marketplace
Concept
Italian AI skill exchange
Ready-made AI skills for Italian paperwork: SDI invoices, PEC, F24, Italian NDAs.

INSIDE THE LAB

Three demos you can click through.

INTERACTIVE MOCKUPS · SAMPLE DATA

COMPLIANCE · REGISTRYDEMO

>PARSING11:02:18
LIMITED-RISK·User-facing disclosure
>NIST AI RMFCLASSIFIED11:00:48
FRAMEWORK·Govern · Map · Measure · Manage
>REF-302CLASSIFIED10:59:18
MINIMAL-RISK·Voluntary code aligned
>ISO 23894:23ARCHIVED10:53:48
RISK MGMT·Risk management guidance
>REF-611ARCHIVED10:52:18
HIGH-RISK SYSTEM·Conformity assessment
>REF-445ARCHIVED10:50:48
LIMITED-RISK·Chatbot disclosure notice
>OECD AIARCHIVED10:49:18
PRINCIPLES·Framework alignment
>REF-901ARCHIVED10:47:48
MINIMAL-RISK·Code of practice aligned
>GDPR Art. 22ARCHIVED10:46:18
AUTOMATED DECISION·Safeguards required
>REF-733ARCHIVED10:44:48
HIGH-RISK SYSTEM·Risk classification updated
>REF-501ARCHIVED10:43:18
LIMITED-RISK·Notification obligation
>NIST 800-218ARCHIVED10:41:48
SECURITY·Supply-chain controls referenced
>ISO 42001:23ARCHIVED10:40:18
MGMT SYSTEM·Governance framework linked
>REF-127ARCHIVED10:38:48
HIGH-RISK SYSTEM·AI Act Art. 14 · human oversight
>REF-389ARCHIVED10:37:18
MINIMAL-RISK·Self-assessment recommended
>ISO 27001:22ARCHIVED10:35:48
COMPLIANCE REF·Controls 8.2 / 8.3 linked
>REF-612ARCHIVED10:34:18
LIMITED-RISK·Transparency obligations only
>REF-843ARCHIVED10:32:48
HIGH-RISK SYSTEM·AI Act Art. 6 · Annex III

AI Compliance

EU AI Act registry (demo)

A new AI system shows up in the company. The demo walks through what a registry would do: read the policies, classify the risk, draft the technical file.

Open the demo below

STAY OS · INBOXDEMO

host

Welcome! Your door code arrives the day before check-in.