Close to the user.
Local inference keeps the task near the interaction. Our browser demos process your input on your device.
ABOUT EMBERMOTE
A little intelligence.
Right where you are.
We’re building small models for the everyday moments in software. A spoken thought. A few words. A shape drawn by hand.
THE IDEA
Useful things.
A smaller footprint.
A model can do one thing well and become a useful part of something much bigger. We’re interested in the small tasks that make an app feel more thoughtful: understanding a message, transcribing a recording, or making sense of a sketch.
Embermote explores how much of that work can happen on the device you already have. We build around clear inputs, compact artifacts, and local inference, with Node.js and Kotlin integrations.
Small, with a purpose.
We compare useful tradeoffs between model size, recognition quality, and runtime. The right footprint depends on the job.
Show the work.
Model pages explain their inputs, sources, licenses, and measured scope. Improvements need evidence, and upstream work deserves credit.
WHERE WE ARE
Early work.
Open possibilities.
Our first collection spans text, audio, and drawn shapes. These are research releases with different levels of maturity, from small experimental classifiers to a compressed speech model with a published evaluation.
We’re developing the models and their integrations together. Each model page is the place to check what works today, what has been measured, and what still needs work.
Read the research status →