Assamese from the first token.

Three dense models, plus voice, documents, and a script‑aware tokenizer.
Deployed where you control it.

See the family → Try the chat

Dense models

Supernova, Axomiya, and Router.

Supernova 28B · multimodal agentic Axomiya 31B · Assamese language Router 248M · instruction-tuned

ASRSpeech recognition

TTSText to speech

Voice AgentSpoken interface

Print AnalyzerPrinted documents

Handwriting AnalyzerHandwritten documents

Axomiya, Router, TTS, and the tokenizer.

Navdyut base models, 15M to 240M, on Hugging Face.

The 28B agentic model, deployed inside your boundary.