
piq
Lets an agent look at a web page, window, or terminal and ask what it shows.
piq gives an agent eyes. Point it at a web page, a terminal program, a window, or an image, and it captures it and (optionally) asks a vision model what it shows, grounded with the page or terminal text when that's available.
It prefers local models when one is running and only falls back to a remote gateway after that. I use it constantly, including to review the screenshots on this site.
$ piq image:$PWD/page.png -m openrouter/google/gemini-3.5-flash -q 'In two sentences: what is this page, and what can a visitor do here?' ● image: tmp1767474 via: file model: openrouter/google/gemini-3.5-flash ● screenshot: file://~/ai/scratch/demo-t/piq/page.png ● pixels sent to: PIQ_GATEWAY (http://localhost:4000) This page is the introduction to "Origami Odyssey," an educational website featuring precise, step-by-step origami folding lessons. Visitors can click the "Open the lessons" button to begin learning how to fold a sailboat, access a user guide, or customize the website's visual theme.