Pillar 04 · AI solutions

AI systems that work inside your business, not in a demo.

We build and operate AI products. Our own, UnoInbox, centralises every customer conversation with an AI agent and a CRM. And when your need is specific, we design custom agents, assistants and pipelines on your data, with the guardrails to run them in production.

  • 1 inbox

    for WhatsApp, Instagram, Messenger and TikTok

  • 24/7

    AI first response in Darija, French, English

  • 100%

    of outbound messages pass a compliance gate

  • OpenAI
  • Anthropic
  • Mistral
  • WhatsApp Business API
  • Instagram API
  • Messenger

What we do

Custom AI solutions

When an off-the-shelf tool does not fit, we build.

  • AI customer agents

    Agents on WhatsApp, Instagram, your website or by voice that answer, qualify, book and escalate. Grounded in your catalogue and policies.

  • Knowledge assistants (RAG)

    Ask questions across years of documents, contracts and procedures and get a sourced answer in seconds, with access rights respected.

  • Internal copilots

    Assistants embedded in the tools your team already uses: drafting, summarising, searching, filling forms.

  • Document processing

    Invoices, orders, claims, IDs, contracts: extraction, validation, routing. Thousands of pages a day without a keyboard.

  • Autonomous workflows

    Multi-step agents that research, decide within rules and act across systems, with human approval on the steps that matter.

  • Web platforms & portals

    Client portals, booking systems and internal apps built around the AI features, when the system needs its own front door.

Featured product

UnoInbox

One inbox for all your messaging channels, an AI agent that answers and qualifies, a CRM that remembers, and WhatsApp campaigns that respect the rules. Built in Morocco for Moroccan SMEs.

  • WhatsApp, Instagram, Messenger today; TikTok coming
  • AI replies with automatic hand-over to a human
  • Contacts, labels, lists, consent ledger, audit log
  • Template campaigns with cost preview and quiet hours
WhatsAppInstagramMessengerTikTok· Coming soon

How we build

Boring, proven technology around the model. The model is the smallest part of a production AI system.

  • Model choice per use case

    OpenAI, Anthropic, Mistral or open models hosted in the EU or on your servers, chosen on accuracy, cost, latency and data sensitivity.

  • Guardrails in code

    Policy gates, escalation rules, rate limits and audit logs are enforced by the system, never left to the prompt.

  • Evaluation before launch

    A test set of real conversations and documents; we measure before we ship and after every change.

  • Observability

    Every decision is logged. You can see what the AI said, why, and what it cost.

Use cases

Where we have deployed

Three sectors, three very different systems.

  • Healthcare

    WhatsApp booking agent

    Bilingual Darija/French agent connected to the clinic agenda. 62% of appointments booked without a human.

  • Distribution

    Order extraction pipeline

    E-mail and PDF orders read by AI, validated against stock, created in the ERP. −73% processing time.

  • Consulting

    Knowledge assistant

    Ten years of deliverables searchable in natural language, with per-client access rights. Answer in 3 seconds, always sourced.

  • Retail

    Instagram & Messenger sales assistant

    Product questions, availability and store hours answered instantly; purchase intent routed to a seller.

  • Real estate

    Lead qualification agent

    Budget, area, timing collected by WhatsApp before an agent calls back. Visits booked directly.

  • Education

    Admissions assistant

    Programme questions, fees and documents handled 24/7 in three languages during enrolment season.

How it works

From use case to a system in production

  1. Use-case definition

    What the system must do, for whom, with what data, and how we will know it works.

  2. Prototype in two weeks

    A working version on a slice of your real data. You test it, we measure it.

  3. Hardening

    Guardrails, integrations, evaluation set, access control, logging. The part that makes it safe to run.

  4. Operate and improve

    Launch, monitoring, monthly review of conversations and costs, continuous tuning.

What you receive

What you receive

  • A production system deployed in your infrastructure or ours
  • Source code and accounts in your name
  • An evaluation report with accuracy on your real cases
  • Documentation and an operations runbook
  • Training for the team who will supervise the AI
  • A monthly review of conversations, costs and improvements

Tools we work with

  • OpenAI
  • Anthropic
  • Mistral
  • WhatsApp Business API
  • Instagram API
  • Messenger
  • PostgreSQL
  • Go
  • SvelteKit
  • Cloudflare

Typical outcomes

Typical outcomes

62%
of appointments booked by the agent

private clinic, Rabat

3 s
to a sourced answer across 10 years of archives

consulting firm

−80%
missed calls

WhatsApp agent taking the front-desk load

24/7
first response time

escalation to humans in office hours

Frequently asked questions

The questions we hear most often. Yours is probably here.

UnoInbox or a custom agent?

If your need is customer conversations on WhatsApp, Instagram or Messenger with a CRM, UnoInbox is faster and cheaper. If the need is internal, document-heavy or deeply integrated, we build custom. We often start with UnoInbox and extend.

Will the AI say something wrong to a customer?

It can, so we design for it: the agent is grounded in your approved content, escalates when unsure, never handles payments or legal commitments, and every conversation is reviewable. Accuracy is measured on your real cases before launch.

Where is the data stored?

In the EU or on your own servers, encrypted at rest. Model providers are chosen per use case and never train on your data. We document the full data flow.

Does it work in Darija?

Yes. Darija, Arabic, French and English, including mixed messages and Arabizi. Escalation keywords are handled in all of them.

Who owns the system?

You do: code, accounts, data. We can operate it for you under a support plan, or hand it over fully.

Next step

Show us the conversations or documents that pile up.

We will tell you in thirty minutes whether UnoInbox fits or whether a custom system makes more sense, and what each would cost.

30 min · video call · French, Arabic or English · no preparation needed