Multilingual LLM 01
Sawalni
Conversations in the languages people actually use.
Answer across text, image, video, and speech.
Darija · Amazigh · Arabic · French
Explore SawalniModels / 01—06
Models
Six connected models for language, search, conversation, and voice. Open a card for recorded outputs, or explore how they work together.
See what the stack can build →The collection
Language → understanding → action
Language identification 02
Gherbal
Know the language before the first reply.
Route text across 214 languages and 28 scripts.
Global coverage · local nuance
Explore GherbalSemantic embeddings 03
Madmon
Find meaning across a language boundary.
Retrieve relevant content across Arabic and English.
Search · retrieval · ranking
Explore MadmonLinguistic word encoder 04
Sawtone
Let a word travel between scripts.
Connect related forms and spellings.
Arabizi · Arabic · Latin · Tifinagh + more
Explore SawtoneSpeech recognition 05
Daktilo
Make a recording searchable.
Turn human speech into a usable transcript.
Darija · Arabic · French · Spanish
Explore DaktiloSpeech synthesis 06
ASAT-v3
Give a written line its own voice.
Create speech for a product or a story.
Darija · Tamazight · multilingual voice
Explore ASAT-v3Open language atlas
Wikilangs.
Move across 360 Wikipedia languages. Explore where they are used, their training groups, real text, and five kinds of open models.
Enter the atlas →Build with the collection
Build with the stack
The Sawalni API gives you live inference, documentation, and a playground for your own inputs.
Explore use cases → Open API Portal