Export a chat

Telegram

In Telegram Desktop (macOS, Windows, Linux): open the chat → ⋯ → Export Chat History… → Export.

• Keep photos, videos, voice messages, stickers and GIFs switched on: the persona uses them.
• Both formats work: HTML (messages.html) and machine-readable JSON (result.json).
• Drop the export folder, or a .zip of it, onto the page.

WhatsApp

On the phone: open the chat → ⋮ → More → Export chat (iPhone: tap the contact name → Export Chat) → Include media. Send the .zip to your computer and drop it onto the page.

A chat exported without media is a single .txt, and that works too.

Large exports

Exports of 100,000+ messages and many gigabytes work the same way: drop the folder. The page shows what is happening:

1. Reading folder… counts the files (the browser can take a moment to list tens of thousands).
2. The chat goes first. The message pages (or result.json, or the WhatsApp .txt) are uploaded and read before any media. If this export is already imported, the import ends here and opens it, and no media is uploaded. If the media won't fit on the disk, you're told how much space is needed before anything big is sent.
3. Uploading 1.2 GB of 3.4 GB (35%) · 9,000 of 30,905 files: the media follow in small batches, one after the other. The browser never builds one giant request, and a batch that breaks off is retried on its own. There is no size limit when you run it on your own computer.
4. Reading messages: page 37 of 121 · 36,000 messages: all pages are read once, in parallel.
5. Checking media and Moving media: files are hashed to find duplicates and moved (not copied twice) into the data folder.
6. Saving, then the New chat dialog opens.

Only one import runs at a time; an error stays visible in the drop zone and you can choose the folder again.

In the desktop app none of this uploading happens: the app reads the folder, .zip or .txt from where it is (steps 4 to 6), which is faster and unaffected by the browser problems below.

When a file can't be uploaded (browser)

A file the browser can no longer read (moved, changed, or an iCloud/OneDrive placeholder that isn't downloaded) is found and skipped, and so is a single file that keeps breaking its batch: the batch is split until that file is isolated. Skipped media shows as missing in the chat; the message after the import names one, and the browser console lists all. When the chat file itself can't be sent, the import stops and names it. Errors say what the server received, e.g. the server received 1,200 of 4,096 bytes, so a connection or browser problem is easy to tell apart.

Nothing leaves your computer during an import and no model is called. Media is described only when you press Describe in the media library.

With rooftop --import <folder> the folder is read in place and its files are copied, never moved; progress is logged. For a huge export on the same computer this is the quickest way: no upload, and no second copy in a temporary folder.

What is understood

Source	Files
Telegram HTML	messages.html, messages2.html, …
Telegram JSON	result.json (used when a folder has no messages.html)
WhatsApp	WhatsApp Chat with <name>.txt (Android) or _chat.txt (iOS), in a folder, a .zip or alone

• Telegram: text with line breaks, links, quotes and code; replies and partial quotes; forwards; edits; service messages; reactions; link previews; photo, video, round video, voice, audio, file, sticker (.webp, .webm, .tgs), GIF, location, contact, poll and call.
• WhatsApp: Android and iOS layouts, 12- and 24-hour clocks and both day/month orders, detected per file. Multi-line messages, edits and attachments are kept. Deleted messages and the encryption notice are skipped, and <Media omitted> is marked as unavailable.

Media files are hashed, so a sticker sent five times becomes one library item (stk_…, gif_…, pho_…).

Reactions in HTML exports

HTML exports don't say who reacted. In one-to-one chats a reaction is attributed to the other participant, and to both when the count is 2.