Anyone who has handed a document to an AI and asked it to summarize, cite, or analyze that document knows the sinking feeling when you can't verify what it's pulling from. A developer published Eyeball, a small open-source tool that shows inline screenshot evidence from source documents directly alongside AI output - so you can see the original passage yourself without switching tabs.
The logic is straightforward: instead of trusting that an AI correctly quoted a PDF or webpage, Eyeball retrieves and displays a screenshot of the original passage next to the claim. The project description puts it plainly - screenshots are harder to hallucinate (when AI models confidently state things that are false) than text. An AI can fabricate a quote, but it can't easily fabricate a matching screenshot from a real document.
The tool is aimed at high-stakes work where errors are expensive: legal document review, medical research, financial analysis, compliance work. The standard workflow in those contexts involves bouncing back and forth between an AI's output and the original source to verify each claim manually. Eyeball tries to collapse that back-and-forth into one view.
This is a GitHub project in early stages, built for people who have already been burned by hallucinations in high-cost contexts. It doesn't prevent hallucinations from happening - it makes them faster to catch. No pricing, no SaaS product. Just a tool you run yourself.