Zet: an open-source layer on Laya that knows when to ask a human

A developer released Zet, an open-source Apache-2.0 layer built on top of the Laya agent framework that uses conformal prediction sets and a Learn-then-Test threshold, calibrated per language, to mark each answer as sure or unsure so automated responses stay within a user-chosen error budget. On the MASSIVE benchmark with a 5% budget, Zet automated 84% of English and 80% of Swedish requests across a 6-scenario task with zero errors in 75 sure English answers, and automated nothing on a harder 18-scenario task while its prediction sets still contained the correct answer 97-99% of the time. The tool runs locally on ONNX Runtime without PyTorch and includes a web interface for task creation, uploading labeled examples, calibration, and review of unsure answers.

Confidence scores aren't error rates. Zet sits on top of Laya and marks every answer sure or unsure, using conformal prediction sets and a Learn-then-Test threshold calibrated per language, so the answers it automates stay within an error budget you choose say 5% . On MASSIVE human-labeled, 300 examples per language, 5% budget , it automated 84% of English and 80% of Swedish requests on a 6-scenario task, with 0 errors in 75 sure English answers. When the task got harder 18 scenarios , it automated nothing, while its prediction sets still contained the right answer 97-99% of the time. That refusal is the feature working. Runs locally on ONNX Runtime with no PyTorch, with a web interface for creating tasks, uploading labeled examples, calibrating, and reviewing unsure answers. Apache-2.0, built on Laya not endorsed by its authors .