The Object Knows

The Object Knows is an agentic investigation layer for the physical world. Most visual AI stops at recognition: it tells you what it thinks it sees. The Object Knows treats an image as the starting evidence for an investigation. Point your camera at something and ask what you want to uncover. GPT-6 Astra combines visual evidence, available photo context and web research to investigate beyond what is visible. It separates what is directly observed, inferred, externally supported and still unknown, building a traceable evidence trail rather than collapsing everything into one confident answer. When the available evidence is insufficient, it identifies the most useful physical observation needed next and asks the user to provide it, allowing the investigation to continue with new evidence. In our demo, an anonymous photo of a coffee contains no visible café branding. The Object Knows uses the available evidence to narrow the investigation to Valletta and independently surfaces LOT61 as a candidate, which was in fact the café where the photo was taken. Crucially, it refuses to treat proximity as proof and keeps the identity unresolved until further evidence can establish it. The camera is only the starting evidence.