TinyDecide
The smallest, ultra-fast decision-making model. Ask it plain-English questions about a message and it answers them all in one pass, offline, on a tiny chip, with no training.
Many apps need to make quick decisions about text. Is this email spam or urgent? Which app should handle this voice command? What time does this message mention? Usually, this means sending the text to a large cloud-based AI model, which adds cost, delay, and a dependency on the internet.
TinyDecide handles these questions directly on the device. You just ask in plain English, no training required. It can pick an option from a list, answer yes or no, rate something on a custom scale (like tone or priority), or extract details like time, amount, name, or place. Because it answers questions in a single pass rather than generating text word-by-word, it is incredibly fast and provides a confidence score for its answers.
At just 6.2 MB, it runs on an ESP32-S3, in a web browser, or in your own JavaScript, Python, or Rust code, without a server or an API key. Pocket Notes uses it to sort notes on a handheld device. You can try it in the browser playground, where it runs on your own machine.