Artificial intelligence in Algeria: what a company can genuinely do with it

In Algeria, the artificial intelligence that is useful in business is neither a robot nor a public chatbot: it is software that reads your own data and answers your teams in French. The uses that are genuinely being deployed are document search, administrative assistance, automatic reading of paper documents, and forecasting stock-outs or breakdowns. Since 30 December 2025, two presidential decrees also govern data governance and information systems security: the question is no longer only technical, it is regulatory.

This guide covers the whole subject: where the country stands, what the law says, who does what in the ecosystem, what works sector by sector, the technologies involved, and how to choose a provider without getting it wrong.

AI in Algeria in 2026: where do things stand?

In a few months the country moved from intentions to a legal framework. Decrees no. 25-320 and no. 25-321, published in the Official Journal on 30 December 2025, organise national data governance and approve an information systems security strategy running to 2029. The State has also invested in research infrastructure: in 2024 CERIST opened a digital innovation facility with computing capacity intended for AI.

On the ground the picture is more nuanced. Most Algerian companies have not yet reached the artificial intelligence stage: they are at the digitisation stage. Their procedures are on paper, their history in scattered spreadsheets, their documents in cabinets. And AI only works on what has been digitised.

This is the most underestimated point of the subject: in Algeria, the first workstream of an AI project is almost always a data workstream. The companies moving fastest are not the ones buying the most powerful model, but the ones that started recording their activity a few years ago.

What does AI actually change in an Algerian company?

It removes retyping. A control report filled in by hand on the shop floor, a prescription, a delivery note: these documents are copied into a spreadsheet by someone today. A vision model reads them and feeds the system directly.

It makes dormant material searchable. Years of procedures, contracts and reports become a base you can question in plain French, with the source cited back.

It anticipates instead of observing. A stock-out, a drifting machine setting, a recurring unpaid invoice: these are patterns the data already contains and that nobody has time to look for.

What it does not do: decide in your place. Software that steers a medical diagnosis or rejects a job application on its own carries a responsibility that neither the law nor common sense leaves to a machine.

The legal framework: which texts apply?

Four texts now structure the use of data and AI in Algeria. The last two are recent and still little discussed.

Text	Subject	What it changes for you
Law no. 18-05 of 10 May 2018	Electronic commerce	Governs online selling and electronic payment
Law no. 18-07 of 10 June 2018, as amended	Protection of personal data	Stronger protection for health data; transfer outside the territory subject to conditions
Presidential decree no. 25-320 of 30 December 2025	National data governance framework	Mandatory classification and cataloguing for entities charged with a public service
Presidential decree no. 25-321 of 30 December 2025	National information systems security strategy 2025-2029	Sets the country's information systems security trajectory

Both decrees appear in the Official Journal no. 87 of 2025, pages 4 and 6.

Decree 25-320 calls for a clarification, because it is often presented too broadly. It does not cover every company: it concerns public institutions and administrations, together with bodies and companies, public or private, charged with a public service (article 3). For those, it requires data to be classified by sensitivity level and catalogued before any exchange. We analyse it article by article in a dedicated article.

One detail of the text deserves the attention of any company choosing software: the national interconnection infrastructure it provides for is built "separately from the internet network" (article 11). A system whose core functions depend on a foreign online service will be hard to connect to it.

Who does what in the ecosystem?

Knowing the institutions tells you whom to address and which rules are coming. Four of them are named directly by decree 25-320.

• The High Commission for Digitisation, created by presidential decree no. 23-314 of 6 September 2023. It draws up the data classification and cataloguing references, then establishes and administers the national interoperability system (decree 25-320, articles 8 and 17).
• The information systems security agency, which rules on the security-level classification of data (article 16).
• The national authority for the protection of personal data, which checks that the classification of personal data complies with the law (article 15).
• The national information systems security council, which rules on the operation of the framework (article 14).

On the innovation and research side:

• ANPT — the national agency for the promotion and development of technology parks — designs the country's technology parks, including the Sidi Abdellah Cyberpark in Algiers.
• CERIST — the research centre for scientific and technical information — founded in 1985 under the Ministry of Higher Education and Scientific Research. In 2024 it opened a digital innovation facility and a hub focused on AI and cybersecurity.
• The Algerian Startup Fund, a public investment fund, engaged with CERIST in a national programme to create technology projects announced in 2025.

Where does your data go when you use AI?

This is the question that decides everything else. With an online service — whichever it is — every question asked and every document uploaded leaves your company, your country, and the legal framework that protects you. For a patient file, a contract or a payslip, that is not a technical detail: it is a transfer of data to a third party.

Law 18-07 makes that transfer outside the territory subject to conditions, and strengthens the protection of health data. A clinic that queries a foreign AI service about its own patient files steps outside that framework without realising it.

The alternative is not to give up on AI but to move it: the model is installed on the company's hardware. It reads the data on site, and nothing leaves the building.

Online service or AI installed on your premises: which to choose?

Criterion	Online AI service	AI installed on your premises
Where your data goes	To foreign servers	It does not leave
Legal framework	Transfer outside the territory to justify	No transfer
National network separate from the internet	Hard to connect	Compatible by design
Cost	Usage-based billing, rising	Hardware once, then nothing
Internet outage	Service unavailable	Keeps working
The model changes without notice	Yes	No, it is yours
Set-up	Immediate	A few weeks

The online service keeps one real advantage: you start in five minutes. It remains the right choice for anything that touches no sensitive data — drafting an advert, translating a public text. As soon as the question concerns your files, the trade-off reverses.

Which uses work, sector by sector?

Healthcare. Search across records and protocols, administrative assistance, management of consumable stock, preparing practitioner pay. Health data being the most protected by law 18-07, this is the sector where local deployment is most clearly required. An example in service at a clinic in Oran.

Industry and energy. Automatic reading of handwritten records, forecasting from production parameters, search across technical documentation. On isolated sites — industrial zones, the southern oilfields — a local model keeps working when the link drops. Where to start in industry.

Retail and distribution. Stock-out forecasting, tracking returns on cash on delivery, order processing. Opening an online shop in Algeria.

Education and training. Enrolment and group management, academic follow-up, search across course materials.

Bodies charged with a public service. Classification and cataloguing of data within the meaning of decree 25-320, document search across internal procedures, with access boundaries enforced by sensitivity level.

Which technologies are involved?

Behind the word "AI", a handful of specific techniques cover most business needs. It is worth knowing them to judge a proposal.

• Language models served locally — open models running on your hardware, which understand and write French.
• Retrieval-augmented generation (RAG) — the model answers from your documents and cites its sources, instead of inventing.
• Automatic document reading — OCR and computer vision, including on degraded handwriting.
• Role-partitioned agents — each user reaches only their own scope of data, enforced before reading and not at display.
• Speech transcription — dictated notes turned into text, on site.

These techniques are not equally mature. Our catalogue shows the real status of each: in service at a client, delivered on assignment, or validated prototype.

How do you choose an AI provider in Algeria?

The market now counts many players with a similar message: sovereignty, local deployment, bespoke AI. To tell them apart, ask questions whose answers can be checked.

Question to ask	Why it matters	A good answer
Which client uses your AI every day?	A demonstration does not prove use	A name, a sector, a case study you can read
Where does the model run?	Sovereignty, law 18-07, decree 25-320	On your servers, with no outbound call
Who owns the code?	Your future dependence	It is handed over to you
How will delivery be validated?	Knowing what is owed	An acceptance test plan and a signed certificate
What is the real status of each product?	Telling what runs from what is planned	A stated status, product by product
How are access rights partitioned?	Classification by sensitivity	Before reading, not at display

We go through each of these questions, and the answers to insist on, in our guide to choosing an AI provider.

How much does an AI project cost?

The cost does not come from the model, which is open and free. It comes from three items: the machine that runs it, the work of adapting it to your data and rules, and training your teams.

A document assistant on an existing collection is sized differently from a complete ERP with embedded AI. We quote after seeing your real data, never from a description — and we say before starting whether the volume you hold is enough for the result to mean anything.

The cost items of a clinic ERP give a first idea of how a budget is structured.

Where to start without getting it wrong?

With a problem that costs money today, not with the technology. Projects that fail start with "we'd like to do AI"; the ones that succeed start with "our nurses spend two hours a day copying out temperature charts".

Next, check that the data exists. Forecasting breakdowns assumes a history of breakdowns; document search assumes digitised documents. It is the first thing we look at, and we sometimes say it is too early.

Finally, start small and verifiable. One module working in one department convinces more than a global project nobody ever sees running.

Training your teams in AI

Software the team cannot use produces no value. We run in-house training, on your premises and on your own data: artificial intelligence in business, writing instructions for language models, data analysis, Power BI and Power Platform. The catalogue holds five courses, that is eighteen days and one hundred and twenty-six hours of programme.

Which business AIs already exist for your sector?

We design one AI per trade rather than a generic assistant, because the rules of a dental practice have nothing in common with those of a pharmaceutical laboratory. See the full range, with the real status of each: deployed at a client, being adapted, or at mock-up stage.

• Healthcare — DentalNova, MedicalNova, BioNova, ImagerieNova
• Pharmacy — PharmaNova, OfficineNova, DistriNova
• Industry — IndusNova, MaintNova, QualiNova
• Education — FormaNova, ScolaNova
• Documents and recruitment — DocuNova, MatchNova

The technical foundation is common to all of them: how we deploy a model on your own servers.

By city

This guide applies across Algeria. Two pages go into detail depending on your situation: in Oran, what is already running at companies here, and in Algiers, how a project run from Oran unfolds. We work on site in ten provinces: Oran, Algiers, Blida, Tipaza, Constantine, Annaba, Sétif, Tlemcen, Sidi Bel Abbès and Hassi Messaoud.