Case study
Sappy, SapinSapin AI's community manager agent, runs on NYO
SapinSapin AI is a community project building open speech and language foundations for Philippine languages. Its assistant, Sappy, answers questions about the project in Tagalog, Cebuano, Ilocano and Hiligaynon - and every answer is written by a model that supports those languages, called through NYO.
“Every voice belongs in the future.”
- sapinsapin.ai
Philippine languages are underrepresented in the data that shapes modern AI. SapinSapin AI - Sariling AI PINas - is an open community project that publishes speech and text datasets, speech recognition and text-to-speech models, and text-generation models for Filipino, Cebuano, Ilocano, Hiligaynon, Waray, Bikol, Kapampangan, Pangasinan and Tausug, all on the Hugging Face Hub. The name comes from sapin-sapin, the layered rice cake: many languages, one country.
Sappy is the project's community manager agent. It welcomes whoever turns up, chats with them, and answers in their own language what the project has, how to use it, and how to take part - so a visitor does not have to read every dataset card, and the volunteers behind the project do not have to answer the same questions by hand.
How Sappy answers
Three layers, like the kakanin: the visitor asks, the project's knowledge base supplies the facts, and NYO supplies the model that writes.
- 01
Ask
A visitor types a question on sapinsapin.ai the way they would say it - in Tagalog, Cebuano, Ilocano or Hiligaynon, with English mixed in.
- 02
Look it up
Sappy searches the project's knowledge base - its datasets, models, cards and write-ups - for the passages that answer the question.
- 03
Answer
A model that supports those languages, called through NYO, writes the answer from those passages in the visitor's language, and the page shows how many it drew on.
Why NYO
A model that supports their languages
Sappy runs on a model that supports Tagalog, Cebuano, Ilocano and Hiligaynon, so people can ask in their own language rather than in English. NYO serves it behind the same endpoint as every other model, so the project can move to a better one as it appears without touching Sappy's code.
Open all the way down
SapinSapin AI exists to make Philippine language technology open and inspectable. Running its assistant on open-weight models, rather than a closed API, keeps the whole stack consistent with that: open data, open speech models, open language models.
One call from a small Worker
Sappy is a small Cloudflare Worker. It calls NYO's OpenAI-compatible endpoint with one API key - no provider accounts, no SDK, no infrastructure to run. Credits are prepaid, so an open research project knows what its assistant can spend.
Stays on the project
Sappy answers about SapinSapin AI and its community, grounded in what the project has published. NYO's guardrails let a team hold an assistant to its purpose and see on its own observability page what was turned away.
Building for a Philippine audience?
A knowledge-grounded assistant in Philippine languages on open models, with one key and guardrails built in.
Sappy is built and run by SapinSapin AI; its images are theirs and used here to describe their project. Sappy can make mistakes - check a dataset or model card before relying on a detail. NYO provides the model access described here and nothing else.