When you search for “artificial intelligence companies in Tunisia,” you almost always find the same thing: a foreign directory, a list of service providers ranked by number of reviews, and very little information about what these companies actually build.
This page answers differently. It lists the players who are doing AI in Tunisia, what they are building, and what is still missing. We update it every quarter.
What Tunisia has already proven
Two exits have changed the way the world looks at Tunisian tech.
InstaDeep, founded in 2014 in Tunis and then established in London, builds decision support systems based on reinforcement learning. BioNTech finalized its acquisition in July 2023 for approximately $549 million. This is to date the largest acquisition of an AI company born in Africa.
Expensya, born in Tunisia and based in Paris, automates expense report management through automatic document reading. Swedish company Medius finalized its acquisition in June 2023.
These two stories say the same thing: the talent is there, and it is producing world-class companies. They also say something more uncomfortable. In both cases, the final value was captured elsewhere.
Hardware
Enova Robotics, founded in 2014 in Sousse, manufactures surveillance and service robots. It is one of the few Tunisian players to work with hardware and not just software, and a useful reminder that artificial intelligence does not stop at a screen.
Language, the true Tunisian territory
This is where Tunisia has something that no one can build for it.
iCompass and InstaDeep published TunBERT in 2021, presented as the first pre-trained language model for the Tunisian dialect, with its code and datasets freely accessible.
Linagora published Labess-7B-Chat, an open model fine-tuned on derja, then a complete speech-to-speech chain in Tunisian: speech recognition, language model, speech synthesis.
ClusterLab works on automatic Arabic processing, starting from the observation that large international models remain weak on dialects and cultural context.
216Bot works on code-switching, this mix of Arabic, French, and English in the same sentence which is the normal way of speaking in Tunisia.
One result connects them all. A research paper published on arXiv in November 2025, How Well Do LLMs Understand Tunisian Arabic?, measures that large language models poorly understand Tunisian Arabic. This is not an opinion, it is a measurement.
The map, at a glance
| Player | Field | Status |
|---|---|---|
| InstaDeep | Decision AI, biology | Acquired by BioNTech, 2023 |
| Expensya | Financial automation | Acquired by Medius, 2023 |
| Enova Robotics | Surveillance robotics | Independent, Sousse |
| iCompass | Language processing, derja | TunBERT, 2021 |
| Linagora | Open models, Tunisian voice | Labess, voice chain |
| ClusterLab | Arabic processing | Active |
| 216Bot | Code-switching | Active |
| IyedeX | Consumer products in derja | 3amAli, online |
What is still missing
A public derja benchmark. No one publishes an open, dated, and reproducible leaderboard that measures what large models actually understand of Tunisian. Without this, each player claims and no one verifies.
Tunisian data. Existing corpora are small, scattered, and rarely designed for today’s spoken Tunisian. This is the bottleneck for everything else.
Products, not demos. Tunisian research on derja is real and published. What remains rare is the transition from code repository to something a Tunisian opens on their phone on a Tuesday evening without knowing what a language model is.
The framework
Tunisia has adopted a national artificial intelligence strategy for 2030, and the Startup Act remains the reference framework for young technology companies, with a label that grants access to a set of tax and foreign exchange advantages. For an AI company launching today in Tunisia, these are the two texts to read before any business plan.
Where IyedeX is located
We are not a laboratory, and we do not claim to be. IyedeX is a Tunisian company that builds artificial intelligence products for Tunisians, in software as in hardware.
Our first product, 3amAli, is an assistant that responds in Tunisian dialect. It is online, free, and usable today. Our bet is simple: Tunisia does not only need models that speak Tunisian, it needs products that Tunisians actually use.
This page is alive
This map is incomplete, necessarily. If you are building AI in Tunisia and are not listed here, or if any information is inaccurate, please write to us and we will correct it.
We update this page every quarter. Next revision: November 2026.