LanguageCC-BY 4.0
Tunisian Darija Corpus
Conversational Tunisian Arabic — transcribed, de-duplicated, dialect-tagged. Train/eval splits ready for fine-tuning.
1.2M rows · JSONL · open
Build · The Data Hubopen · sovereign · free forever
The infrastructure layer.
Arabic corpora, Tunisian dialect datasets, and open APIs — built sovereign, in Tunis. The data and tools that train the next generation of Arabic-native AI, free for any builder who needs them. No paywalls, no extraction, no foreign custody of our own language.
11M+
Arabic rows
6
Open datasets
100%
Open license
0
Paywalls
Datasets · Sovereign corpora
Six corpora. All open.
LanguageCC0
Classical Arabic Text Set
Fully vocalized fuṣḥā from heritage manuscripts and modern presses. Diacritic-complete, cleaned, paragraph-aligned.
640K docs · Parquet · open
EmbeddingsCC-BY 4.0
MENA News Embeddings
Pre-computed 1024-dim vectors over a decade of Arabic and French regional press. Indexed and ready for retrieval.
3.1M vectors · NPY · open
SpeechCC-BY-NC 4.0
Tunisian Dialect Speech
Native Tunisian speech with verbatim transcripts and speaker metadata. Balanced across age, gender, and region.
420 hrs · WAV + JSON · open
ParallelMIT
French–Arabic Parallel Pairs
Sentence-aligned translation pairs across legal, civic, and editorial domains. Human-verified and deduplicated.
2.8M pairs · TSV · open
VisionCC-BY 4.0
Maghrebi Handwriting
Scanned Maghrebi Arabic handwriting with character and word-level annotations. Augmented for OCR and HTR models.
190K samples · PNG + COCO · open
API & Tools · Open by default
Tools that speak our language.
REST · trilingual
Embeddings API
One endpoint for Arabic, French, and Tunisian Darija vectors. Diacritic-aware, dialect-aware, pinned to Tunis.
REST · MT
Translate API
Bidirectional FR ⇄ AR ⇄ Darija translation tuned on the parallel corpus. Domain hints for legal and civic text.
Python · pip
Dialect Tokenizer
pip install myn-tokenizer. Splits Tunisian Darija and fuṣḥā correctly where generic tokenizers break. Diacritic-safe.