GitHub radar
Classify any text with five questions at once
Convai Innovations released an open-source text classifier. It takes a piece of text and a list of typed questions, then answers all of them at once with confidence percentages in around 33 milliseconds.
Laya is an open-source text classifier from convaiinnovations on Hugging Face. You give it a piece of text and a list of typed questions in three formats: choice between options, yes/no, or a numeric score. It returns all answers at once with confidence percentages. The base model has 421 million parameters. A multilingual variant covers more than 100 languages. Convai built Laya as a direct open alternative to TypeSafe Jev, a paid decision API. According to the published benchmark, Laya is 7.8 times faster on a single question and has 3 times better calibration accuracy. The weights and code are Apache-licensed. You can self-host the model or use the free demo space on Hugging Face without any registration.
Why a vibe-coder should care
If you build a product where text comes in and needs to be routed or filtered, Laya removes the need to call an LLM for every message. One request handles five questions at once, which saves time and API cost compared to sending each question as a separate ChatGPT request. The demo on Hugging Face works in the browser without any setup or registration.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Open this demo space: https://huggingface.co/spaces/convaiinnovations/laya-demo — paste a text like 'My name is Alex, I am interested in your service, please call me.' and write five yes/no or choice questions about it, such as: is this a hot lead? is this spam? should I call them? should I delete this? is a sale possible? Show me the answers with confidence scores.
No installation needed — runs in the browser.
Open on Hugging Face▌ More finds