Grounded intelligence for the real world.

Software assumes. Reality verifies.

Intelligence is breaking out of text. It is learning to act: to run experiments, to simulate, to be tested against reality and corrected by it.

That is only half the picture. The other half is grounding: putting intelligence in contact with the real. What works in simulation is a theory; what works in the field is a solution.

Progress is fastest where a claim can be checked. Most of the real world is not like that yet: its loops produce weak, noisy or unused data. We build the missing half: the environment where a model's claim can be tested, and the models that learn in it.

We refuse the customer's chair.

The story of this new age is told from a few cities. A handful of labs race to build minds that can do almost any kind of work, and the rest of us are invited to rent those minds by the hour. We do not accept that the customer's chair is the only one left for us.

Morocco has spent a generation as a country that executes but does not conceive: others bring the designs, we supply the factory and the hands. Now the same pattern is arriving in AI. A sovereignty you cannot open and repair is just a costume. Better to own a hard system than to rent an easy one that can be switched off from far away. We would rather build it ourselves, clumsily at first, than rent it forever.

So our ambition is not a product. It is a layer of grounded, sovereign intelligence over Morocco's real economy, from bits to atoms: its language, its companies and its machines, owned from the metal to the model.

How we choose problems

We look for real-world loops that produce weak, noisy or unused data, and ask how to make them verifiable. Four questions run through that work.

1. How do you turn the noise of the real world into knowledge stable enough to act on?
2. How do you test a model's claim against reality, instead of against text?
3. What should a person define, and what should they still check, once a loop runs on its own?
4. How much of this can run on hardware you own, and can open and repair?

What we are working on

Four directions, one method. Each is a loop we are working to close.

• LanguageDarija is spoken by nine in ten Moroccans and badly served by general-purpose speech models. Sibawayh, our first model, is speech recognition built for it, with its own evaluation.Sibawayh research log
• EnterpriseAI that runs inside an organisation, on its own infrastructure, and keeps the source of every figure.Asayl AI PlatformMasnaa OS
• EngineeringHow machines are designed and how they are diagnosed. We turn simulation and repair into environments where a model's proposal can be checked before anyone acts on it.
• ScienceExperimental loops in the life and earth sciences: design, make, test, learn, where each experiment produces data that exists nowhere else.

Who we are

Asayl comes from al-sayl (السَّيْل), the torrent. Scattered drops die in the sand; gathered, they become a current that cannot be refused.

Asayl Labs was founded in 2026 by two engineers from ENSA Berrechid. Between us we have built Sibawayh end to end, the Asayl AI Platform and Masnaa OS.

• Elwalid AboulaakoulCofounder · ML research and systems
• Ahmed Ben AhmedCofounder · data engineering and software systems

Academic guidance

We benefit from the guidance of professors working across engineering, infrastructure, applied systems, mathematics and artificial intelligence.

• Prof. Abdeljalil SakatComputer science, mathematics, PINNs, optimization, and numerical modeling
• Prof. Driss HarriziCompetitive intelligence, governance, strategic management, and higher education
• Prof. Ahmed NafidiStatistics, probability, stochastic processes, and applied mathematics
• Prof. Abdellah JamaliComputer networks, cloud computing, AI, deep learning, and network security
• Prof. Moumoun LahcenArtificial intelligence, computer vision, machine learning, and signal/image processing

The door is open

Morocco's engineers today are drops scattered across other people's labs. If you would rather own a hard thing than rent an easy one, write to us.

We work with companies and institutions through pilots that run on their own infrastructure, with their own data.

We are open to research collaborations: shared experiments, datasets and evaluations. A first conversation can start from a question, a sample, a synthetic dataset or a controlled context.

contact@asayl.co

وأَن لَيسَ لِلإنسانِ إلّا ما سَعى

Sa'yuna, our founding essay →

Why we exist, and where the name comes from.