Origain — prompt AI from any page

Compose a prompt from the page you are on, put it in a conversation you already have open, and remember where it came from.

Your prompts have an origin. Origain remembers it. Every chat application files your thinking by time, under a title it generated for you. That is why nobody ever finds anything in one again. The better index is the page you were looking at when you asked — and nothing can build that index except something standing in both places. Compose from the page. Write what you want, then point at what it is about: highlight something, take the page's text, or ask a code editor on the page for its file. Each part gets a label, so the model can tell the problem from the code from your attempt. Put it where you are already working. Origain drops the finished prompt into an AI conversation you already have open — the one with your context in it — instead of starting a fresh chat that knows nothing. Nine assistants are supported, and the panel names them. Or copy it and paste it anywhere. Start from something you have used. The structure of a prompt is the part worth keeping: the instruction and the labels are the same on every page you apply them to, and only the contents change. Origain offers the shapes you have used before and fills them in from the page in front of you — including asking the code editor directly, which is the only way to get a file out of an editor that renders a few lines at a time. It uses what a site already offers. More and more pages publish a clean copy of themselves for language models — a Copy for LLM button, or a Markdown version of the page. Where one exists, Origain says so and gives you a labelled place to put it. You press the site's button; Origain never does. It reads only what you point at, and takes only what you hand it. And it remembers where you asked from. Come back to a page and Origain lists what you asked there, with the way back to the conversation holding the answer — and, if you want it, the answer itself, collected from that conversation and kept beside the question. Then search all of it, across every model you use. One page holds every prompt you have composed and every answer you collected, searchable by words, by site, or by the shape of the prompt. You cannot search Claude, ChatGPT and Gemini together anywhere else; here they are one list, on your own machine, and anything in it can be deleted in a press. Pass a good one on. The shape of a prompt — the instruction and the labels, and where each value was found — copies to your clipboard as a few readable lines, so you can send it to somebody in a message. Paste one you were sent and it is offered on the pages it was made for. No registry, no directory, nothing uploaded or downloaded: a shape carries no contents, no pages you visited and no conversations, and one you paste in brings labels only — every read is still a press, on a site you allowed. What it does not do - It has no server, and makes no network requests of any kind. - It installs able to reach no website at all. Every site, including every AI provider, is allowed by you, one at a time, and revocable one at a time. - It never carries text out of one conversation into another, and never reads a conversation it did not put a prompt into. It can collect the answer to a prompt it sent, on a press, and only when that answer is the answer to that prompt. - It reads a page only when you press a button, and only on sites you have allowed. - No analytics, no tracking, no account, no remotely hosted code. Everything it stores stays on your computer.