An AI company from Tunis
Hannibal answers in Derja.Fennec hears you, Hannibal answers in Derja, Dido reads it back.Fennec hears you and Hannibal answers in Derja.Hannibal answers in Derja, and Dido reads it back.Hannibal answers in Derja.
The lineup
Three models, one language.
Hannibal
Reads and writes Tunisian Arabic, and can look at an image or a video.
Read moreFennec
Small enough to run anywhere, at 23% word error and 9% character error.
Read moreDido
A Tunisian voice, not a Tunisian accent on someone else's.
Read moreFor developers
Build with the API.
All three are live behind one OpenAI-shaped endpoint: Hannibal for chat, Fennec for transcription, Dido for speech. Sign in, create a key, and your first request is the one below.
curl https://baarcha.tn/api/v1/chat/completions \
-H "Authorization: Bearer pk_live_…" \
-H "Content-Type: application/json" \
-d '{
"model": "hannibal",
"messages": [{"role": "user", "content": "chnwa a7welek?"}]
}'Why this exists
A language you cannot buy data for.
Derja is Arabic script, Latin Tunizi and French in one sentence, with no standard spelling and no corpus at any price. So we built the data engine first, and the models second.
The last step of that engine is people. In the game you hear one line a model wrote and say whether a Tunisian would really say it. What enough strangers agree on becomes training data, and what they reject gets rewritten.
9%
character error, on clips nobody trained on
2,000+ h
of Tunisian audio, none of it for sale
13M
speakers, no model that hears them
The names
All of it is Carthaginian.
- Hannibal
- The general.
- Dido
- The queen who founded Carthage.
- Fennec
- The desert fox.