Assay: Deep Dive for AI Chats

Reading an AI answer is easy. Thinking with it is the hard part. Every AI chat interface is built around one verb: prompt. But when you actually think with a model, most of the work is editorial — reading a long answer and deciding which three sentences deserve a second look. That step has no interface. It is copy, paste, retype. Assay gives it one. While you read, tap any word to highlight it. Tap the highlight to widen it — word, sentence, whole paragraph. Tap other words in the same reply to grow the selection toward them, across paragraphs. Collect as many passages as you like, add a few words of annotation to each, and send the whole set back as one prompt. No boilerplate is added: your selection is the signal, and your notes carry the intent. What you get Tap a word in a reply → it highlights, with a small action bar Tap the highlight to widen: word → sentence → paragraph → back Tap nearby words to grow the selection, across paragraph breaks ＋ Add collects it; ✎ Note annotates it, before or after the quote Six rotating highlight colours, so you can see what you already took A tray holding everything you kept, while the chat stays scrollable ↗ To composer writes your fragments and notes into the message box ⭳ Export the whole conversation as Markdown or plain text, your fragments and notes appended — a verbatim local record, named after the chat Every conversation keeps its own list Private by design No account. No server. No analytics. No API keys — your existing ChatGPT or Claude subscription does all the inference. Fragments live in your browser, on your device. The only thing that ever leaves it is the message you choose to send. Open source under MIT: github.com/puj/Assay Works with ChatGPT (chatgpt.com) and Claude (claude.ai). On Android, run it in Firefox. A Project Nothing experiment — this is an experiment, and it might be gone next month. Assay is not affiliated with OpenAI or Anthropic.