Omneity Labs / Overview
Moroccan GenAI.
Global research.
We build a sovereign Moroccan GenAI stack that people can use in products today. We also release data and language models that help the wider world build multilingual AI.
01 / Moroccan GenAI stack
A complete path from input to action.
Text, speech, images, and video frames can enter the same application. Sawalni carries the conversation. Supporting models identify language, connect scripts, retrieve knowledge, transcribe, and speak.
Explain what happens in this video.
Three video frames · input to Sawalniفالفيديو، واحد السيد واقف فواحد السوق، كيشوف فواحد الحيط عامر بحوايج ديال الجلد، قرون، و حوايج خرين. ف اللقطة اللولة، كيدوز يدو على شي حاجة ديال الجلد. ف اللقطة التانية، كيهز واحد القرن كبير. ف اللقطة التالتة، كيشوف مزيان ف القرن و كيتحرك شوية.
Recorded Sawalni API output · source video: IMAD EL MAADI / Pexels
In a product
Sawalni Batbot.
Add a knowledge assistant to a website. Visitors ask questions. Batbot retrieves your pages and documents, then answers with sources.
Explore Batbot →Find the right service.
Hours, appointments, and request tracking. Ask in the language that works for you.
Recorded outputs
Watch the model work.
كتب ليا دعوة قصيرة بالدارجة لورشة ديال الرسم فالرباط نهار الأحد مع 11 الصباح. الأطفال مرحب بهم. جوج جمل فقط.
مرحباً بكم ف ورشة الرسم ديالنا ف الرباط نهار الأحد مع 11 ديال الصباح. الأطفال مرحب بهم.
تبدل الوقت والبلاصة: الورشة ولات ف سلا مع 14:00. عاود الدعوة وخلي الأطفال مرحب بهم، فجوج جمل.
مرحبا بكم ف ورشة الرسم ديالنا ف سلا نهار الأحد مع 14:00. الأطفال مرحب بهم.
02 / Global multilingual research
Language infrastructure for everyone.
The same work that makes our Moroccan stack possible extends to hundreds of languages. We publish data, browser-ready models, language identification research, and linguistic representations so more languages can be part of AI.
Wikilangs
Basic models for hundreds of Wikipedia languages. Explore families, geography, real text, and models that run in the browser.
Wikipedia Monthly
Fresh multilingual Wikipedia data for model builders, including languages with little accessible NLP data.
Gherbal
Identify language and script across diverse web text. The pipeline helps sort multilingual data, including low-resource languages and dialects.
Sawtone
A word-level foundation for 400+ languages. It connects writing, sound, and meaning across scripts for search and language tools.
Wikipedia Labels, spaCy NLP, and Mini Agent extend the research.
About the lab
Built in Salé and Berlin.
Omneity Labs builds language AI products and publishes the research behind them. We start with languages and people that mainstream AI often overlooks, then make the work useful far beyond one region.
Meet the lab →Contact
Build or research with us.
Tell us what you are building or studying. We read messages through the contact form, and you can email us directly.