ChatGPT Export to Markdown - Local Backup that Checks for Missing Messages

Save the ChatGPT chat as Markdown, PDF or .html with images and diagrams; gaps are named, not hidden. Free, unlimited, no server.

Export the ChatGPT conversation you are looking at into Markdown, PDF or .html, with TXT / CSV / JSON copies on top; images, code blocks, tables, formulas and diagrams come along too, and everything lands on your own computer. It also checks the backup it just made: two lines of readings appear the moment the scan ends — "Messages — 18 (you 9 · ChatGPT 9) · 3 shorter than the page" and "Rich content — code 2/2 · tables 1/1 · images 4/4 · formulas 1/2". It never just says "export succeeded"; it shows you every count. And it packs a handoff .zip: this conversation turned into material the next AI can keep working on — Claude, DeepSeek, Kimi, Cursor, Obsidian — without you retelling it. WHY THIS ONE • Free, unlimited — one chat, one Markdown file; downloading and printing have no paywall and no daily quota. Getting your own conversations back should not be a paid feature. • Your data never leaves this computer — no server, no analytics, no login, no token reading, no background requests; the extension's only permission is storage. • The PDF is made here too — it goes through your browser's own print dialog where you pick "Save as PDF", with no sign-up or login just to get a PDF. Want a page you can just double-click, with the images already inside it? Use "Download .html" — it is the same page the print dialog gets. WHAT MAKES IT DIFFERENT ChatGPT truncates and loses conversations, and the official export takes days by email. This extension backs up the conversation you are looking at, right now: it scrolls the whole thread, exports it to Markdown, and saves it on your own computer. The difference is that the backup can be reconciled. Before and after exporting it counts your messages, the characters in each one, and the code blocks, tables, images, links, formulas and diagrams on the page — then shows you the numbers side by side. If something is missing it names it: which messages came out shorter than the page, which formula has no LaTeX source. You never get a silent, half-empty file — and you don't have to re-check each export against the page by hand. FEATURES • Three layers of checks — message counts, per-message character counts, then rich content item by item (code / tables / images / links / formulas / diagrams / writing blocks). Every row in the per-message readings can be copied on its own, so grabbing one answer does not mean downloading the whole file. • It keeps going after the export — click "Hand off: Handoff pack .zip" and you get material for the next AI, not another archive: the full context, an opening instruction, a memory sheet and one tagged note. Hand it to any AI that reads text — Claude, DeepSeek, Kimi, Cursor, Obsidian — and it continues the work instead of asking you to retell it. The pack states which images, formulas and branches did not make it, so the next AI does not assume nothing is missing. The memory sheet is a blank template (what you are doing / what is settled / what must not change / next step), so the next AI asks you those four things. • Long chats are not cut off — ChatGPT renders only a dozen messages per screen, which is why some exporters give you 8 of 19. It scrolls the conversation itself and unions every message it has seen; the scan ends when two passes in a row find nothing new. At the 90-second ceiling it still gives you what it collected, and the progress bar, pass number and count stay on screen. • Images are saved, not dropped — generated images are written next to the file (assets/gen-1.png) inside a ZIP, so the Markdown renders offline and in Obsidian. • Diagrams keep their source — a mermaid chart on the page is an inline SVG, so a text-based export simply cannot see it. We take the chart source the page itself keeps and write it as a ```mermaid fence, which Obsidian and GitHub redraw and let you keep editing. A chart whose source is not on the page falls back to the text drawn on it, and the reading says so: "diagrams 2/3 (1 kept as on-diagram text)". • Writing blocks keep their source — ChatGPT stores the original Markdown of a writing block on the page; we take that verbatim instead of guessing from the rendered HTML. • Local version history — every backup keeps a version, and a version with fewer messages than the last one is flagged in red with both counts and the time of the previous backup, naming which messages no longer appear. Records can be searched by title, filtered down to the ones with gaps, and deleted. • It opens with a table of contents — the file lists each question as "01 · You · Write a story in Canvas…", and pasted into Obsidian / GitHub / VS Code those entries jump straight to the question. No scrolling through a fifty-message chat. • The export looks like the chat — in the PDF and the .html your messages sit on the right inside bubbles and ChatGPT on the left, so the position says who spoke before you read a label; in the .md your message is a quoted block and the reply is plain text. • Export only the messages you want — tick just the ones you need in the panel. The selected range still gets checked word-for-word and item by item, and the header states it plainly: "Scope: partial export, 5 of 23 messages in this chat". Partial files are kept out of the version history so they cannot become tomorrow's baseline and produce a fake "18 fewer than last time". • Nothing of ours ends up in your file — no watermark, no link back to our site, no hidden usage counter. The print page does not lock a font, so Chinese / Japanese / Korean text renders with the faces already on your machine. HOW TO USE 1. Open a ChatGPT conversation and click "Back up this chat" on the right edge (or press Alt+B) 2. Wait for the scan — two lines of readings appear on the panel 3. Click "Download .md"; when the chat has images or writing blocks it is ".zip", and it contains an .html you can just open 4. Want only some of the messages? Expand "Choose which messages to export", tick them, then download 5. Want a PDF? Click "Save as PDF" and pick "Save as PDF" as the destination 6. To let another AI continue this conversation, click "Hand off: Handoff pack .zip" and read 00-START-HERE.md inside it first KNOWN LIMITS, STATED UP FRONT • Code-block language tags — ChatGPT does not mark the language on its code blocks today, so the fence is left empty. • Formulas — a rendered formula without its LaTeX source comes out as rendered text in the body; that is reported as its own gap, not hidden. • Diagrams — a chart whose source is not on the page is saved as the text drawn on it, not as the picture, and the reading states how many. • Empty replies — when the reply was already empty on the page (ChatGPT does this occasionally), the file keeps only "(empty)". That is reported as its own gap rather than passed over in silence; ask for a regeneration, then back up again and it fills in. • Not converged — on a very long chat that did not reach the top within 90 seconds the status says "not converged (earlier messages may exist)" instead of pretending it finished. • Handoff pack — the next AI receives text only: images and formulas arrive as text. DATA AND RETENTION • Data stays in this browser (chrome.storage.local). Uninstalling the extension deletes the backup records too. DISCLAIMER • "ChatGPT" is a trademark of OpenAI. This extension is independently developed: it is not affiliated with, authorized by, or endorsed by OpenAI. • Please back up only your own conversations; exported content remains the property of its original author.