PsiGuard watches the structure of a thing while it moves, not the facts it lands on. Point it at anything that holds together until it doesn’t, and it still has something to say.
Product behavior. The same engine and settings as the API, watching real AI generation over time. Exploratory research. The same signals pointed at something that is not an AI answer. Interesting, published as is, not a product claim.
One long conversation, run continuously and never reset, to see whether PsiGuard keeps a session on track however long it gets.
Open Waypoint
A live AI agent, watched post by post on Moltbook.
Open TinySpark
An AI agent works a long job on its own, step by step, while the job keeps changing under it. PsiGuard watches every step for the moment its thinking starts to come apart.
Open Threadline