Jayce's Lab

Home-lab AI chat

A self-hosted chat with cited web search, images, automatic long-form continuation and answer-quality grading.

What it does

Safety

Quality

Every answer is checked against rules (refusals, format, language, repetition, citations, …) the moment it finishes. Suspicious answers, 👎-rated answers and a random 20% are graded by a different model, and a nightly regression run feeds per-model scores into a dashboard.