An Algerian AI company. We advise on AI, build it, put it into the systems you already run, and ship our own products. Based in Algiers, working across Africa, with clients in Sweden, France and the United States.
Most businesses here are told two things about AI, and neither one helps. The first is that AI changes everything, usually by someone selling a subscription to a tool built for a market that is not this one. The second is that AI is not ready, usually by someone who tried a general assistant on a local problem and watched it invent an answer.
Both are wrong for the same reason. A model that has never seen Algerian notarial procedure, your invoice history, or how your shop actually takes orders will produce something that looks right and is not. Closing that gap is engineering work, and it is done close to the people who will use the result.
That is the job. We are the people who close the gap.
Most clients need one of these. Some move through several over time.
You know AI is relevant and you do not know where. We look at how the work is done today, where time and errors actually go, and what your data can support.
You get a plan with a first project, a cost range, and an honest list of what is not worth doing. No build required. Sometimes the answer is to fix how records are captured first, and we will say so.
A tool that does not exist yet. We scope it, build it, and hand it over: web application, assistant, agent, internal tool, customer-facing product.
You own the code, the prompts, the artefacts and the data at handover.
AI put into the systems you already run, so the output lands where the work happens. Your stock system, your invoicing, your CRM, WhatsApp, whatever is actually in use.
This is usually where projects get hard, because integrations depend on systems nobody in the room controls.
Your developers can build it, they have just not built this kind of thing before. We set the architecture, review the approach, and stay available while they ship.
Cheaper than outsourcing, and the capability stays inside your team.
Everything below is in Kateb somewhere. Client work benefits from that, and the product benefits from the breadth that client work brings.
We index your documents and retrieve the relevant passages at query time, then require every answer to cite what it was drawn from.
Fine-tuning on your data where it genuinely helps, and saying so when retrieval or cleaner data would do the same job for less.
Multi-step work rather than single answers: read, extract, check against a rule, write into a system, flag what needs a human.
OCR, layout parsing and structured extraction from scans, PDFs and photographs. Bilingual by default, because Algerian documents mix Arabic and French.
Answering from your own material rather than general knowledge, in the languages your customers actually write in, darija included.
How you know the thing still works next month. Test sets, measured accuracy on the cases that matter, alerts when quality drifts.
The application around the model. An AI feature nobody can reach is not a product.
A build has an end. You get a scope, a quote, a delivery and a handover. Maintenance afterwards is a separate and explicit agreement, not a renewal running quietly in the background.
Code, prompts, artefacts and data are yours on completion. Documentation is a deliverable rather than a favour, so another developer can pick the work up without us.
Data condition is the largest single cost driver on almost every project. A quote given without seeing it is a guess.
For work carrying professional or legal consequences we build retrieval, citation verification, and a designed path for the system to refuse rather than guess.
Slow connections, mobile data, bilingual documents, right-to-left layout. These are not edge cases in Algeria, they are the normal case.
Turning down a project costs us one invoice. Delivering one that gets abandoned costs the relationship and every referral behind it.
If a spreadsheet or a phone number does the job, we say so during scoping, before you have spent anything.
A number without a test set and a measurement method is marketing.
Refusal is a feature. A confident wrong answer is worse than no answer.
Recurring billing for a finished build creates the wrong incentive on both sides.
The interface translates. The rules underneath do not.
It is where our product lives, where we know the rules, and where being local is the whole advantage.
With local partners, and only where there is a real user on the other side rather than a slide about addressable market.
Custom builds and advisory work delivered remotely, in English and French, scoped in writing and reviewed on working software.
Kateb is valuable because it encodes Algerian notarial law, procedure and fee scales. That is also exactly why it cannot be exported by translating the interface. Rebuilding that substance is most of the work, which is why "AI for Africa" describes a market rather than a product.
The practice is portable, the corpus is not. Advisory work and custom builds go anywhere. Products stay close to the rules they encode until we rebuild them for somewhere new.
Kateb is an AI legal assistant for Algerian notaires, built and maintained by Soukely. It covers conversational research over Algerian legal texts, drafting of notarial acts, document checklists, dossier tracking and fee calculation.
It exists because a general assistant fails at exactly this work. Ask one for a specific article of the Code civil and it will produce something plausibly formatted that may not exist. Where a citation carries professional liability, plausible is not good enough. Kateb retrieves the source text first, answers only from what it retrieved, verifies every reference before displaying it, and refuses when there is no basis.
In beta, currently free, and used by practising notaires. You can try it before you hire us.
A pure product company in a market this size bets everything on one guess. A pure services company starts every year at zero. We run both deliberately, and each one pays for something the other cannot.
Tools we build, own and maintain. Kateb is the first. More are planned for other Algerian professions where the same pattern holds: codified local rules, repetitive document work, and no tool built for the specifics. The product forces a standard of reliability that project work alone would not demand, and it compounds.
Advisory, custom builds and integrations, one business or one person at a time. It shows us what businesses actually need before we commit years to a product, and it covers the runway. Read more on why we run both.
You do not need to be a company to hire us. A shop that wants a chatbot answering customers properly and a firm automating hundreds of documents a day are different jobs, and we size the work to whoever is in front of us.
We work with clients wherever they are. Alongside the Algerian work we have delivered for clients in Sweden, France and the United States, remotely, in English and French.
Soukely is an AI consultancy and product company based in Algiers. We advise businesses on where AI is worth using, build AI tools from scratch, integrate AI into systems a client already runs, and guide in-house development teams building AI for the first time. We also build and maintain our own products, starting with Kateb, an AI legal assistant for Algerian notaires.
We are registered in Algiers as SOUKELY EURL, and Algeria is our deepest market. We are expanding across Africa country by country with local partners, and we have delivered custom builds and advisory work for clients in Sweden, France and the United States, run remotely in English and French.
Both. Advisory work is a service on its own: we review how the work is done today, where time and errors go, and what your data can support, then give you a plan with a first project and a cost range. No build is required, and sometimes the honest answer is that you should fix how records are captured before any AI is added.
It is quoted per project rather than sold as a monthly subscription, because a build has an end. The four things that move a price are the condition of your data, the number of systems involved, how low the acceptable error rate has to be, and who maintains the result afterwards. We look at a real sample of your data before quoting.
Both. You do not need to be a company to hire us. If the job is well defined, the size of the client is not the deciding factor.
You do. Code, prompts, artefacts and data belong to the client at handover, and documentation is a deliverable, so another developer can take the work over without us.
Tell us what you are trying to fix, in plain words. If AI is the wrong tool for it, we will say so. We reply within two working days.