Open a ChatGPT, Claude or Grok data export and search every conversation in it. Your browser reads the file from your disk. Nothing is uploaded, and the page is not allowed to make network connections.
Drop your export here
The zip as you downloaded it, or the conversations JSON file from inside it. You can add more than one.
An export is not a transcript. ChatGPT stores each conversation as a tree that still holds the drafts you edited or regenerated, Claude stores typed blocks where most of the file is not conversation text, and Grok writes the assistant's name three different ways. The reader shows one chain per conversation, from the last message back to the first, with your words and the assistant's words labelled. Those rules come from our own exports, described in the three formats compared.
Search is a plain text match across titles and messages. Tick Only what I wrote to search your own words and skip the replies.
Each assistant has its own steps: ChatGPT, Claude and Grok. If the ChatGPT export never arrives or looks incomplete, see what to check before you retry. The full set of guides is in the AI chat history guide.
Want your assistants to search this history for you?
This page reads an export while the tab is open. ModelBrain keeps the same export indexed on your computer, so Claude, ChatGPT and Cursor can look things up in it when you ask. It reads the same files: ChatGPT export, Claude export and Grok export.
Download ModelBrain (beta)No. Your browser reads the file from your disk. The page is served with a security policy of connect-src 'none', which tells the browser to refuse any network connection from the page's scripts, and the reader is one unminified file, /export-reader.js, that you can read. You can also load the page, turn off your network connection, and then add your file.
A ChatGPT export zip or its conversations JSON files, a Claude conversations-000.zip or its conversations.json, and a Grok export zip or its prod-grok-backend.json. You can add more than one, and they are searched together.
An export keeps drafts you edited or regenerated. The reader follows one chain, from the last message back to the first, which is the conversation you actually read. It also leaves out reasoning traces, tool calls and system messages.
Any single JSON file up to 500 MB. A browser tab cannot hold more than that in one piece. ChatGPT splits its export into numbered files that are well under that size. A very large Claude conversations.json can be over it, and the ModelBrain desktop app reads those as a stream.
No. Nothing is written to browser storage. Closing or reloading the tab clears everything.
Zip files need a current browser: Chrome or Edge 103, Safari 16.4 or Firefox 113, or later. In an older browser, unzip the export and add the JSON file. Found a file it reads wrongly? Write to hello@modelbrain.net.