Point at the exact spot. Say what you think. Hand it to your AI.
πAnnotate is a reading tool for people who read a lot and work with AI. You click where you mean in a PDF, write your remark, and copy it all — the page, the very line under each heart, your remark — into ChatGPT, Claude or any AI chat. No more “page 4, second paragraph, the line that starts with…”.
Drop a PDF onto the page, or click Open. Markdown, PNG and JPEG work too. The document opens in your browser: nothing is uploaded.
A numbered heart drops on the spot, and its [n] appears at the end of your note. Drag a heart to move it; double-click it to remove it.
After its [n]. Markdown and LaTeX are welcome: $p<0.05$ shows as math.
One click copies a plain text your AI can act on: the document’s name, then each [n] with its page, the line of text under the heart, and your remark. In any AI chat, attach the PDF and paste it. On a picture, copy the picture with its hearts too.
Your work stays in your browser as you go. Save downloads one small .json file: drop it back onto πAnnotate, with its document, to continue. AI agents that read files can read it directly.
What your AI receives
This is what Copy for AI gives for the sample paper on the first screen (“Try a sample paper”), word for word.
My annotations on “sample-paper.pdf”, made with πAnnotate. Each [n] is a mark I placed at an exact spot in the document; the quoted text is what lies under it. General note: Reviewing this sample before the reading group. [1] Page 1 — “T = n · (f + w)” Is $w$ really independent of $n$? Long reviews get tired. [2] Page 1 — no text under this mark (a figure, a formula drawn as a picture, or a scanned page), 36% across and 62% down the page Label the y-axis: seconds per remark? [3] Page 1 — “end, the gain doubles: the AI no longer has to guess where a remark applies, because the quote” Overclaimed: “doubles” needs a number.
Where a heart sits on a figure, there is no text to quote: πAnnotate says so, and gives the position.
Good to know
πAnnotate works in your browser, on a computer. On a phone, the page explains it and lets you send yourself the link.
It comes from the PDF’s own text layer. A scanned page has none, and a formula often comes out garbled: then the position is what your AI gets.
Never uploaded, never stored anywhere but your own computer. The details are on the privacy page.