Docs / Connect / ChatGPT export

Index your ChatGPT export in ModelBrain

A ChatGPT export keeps every branch you ever tried, but only the branch you actually read is worth remembering. The chatgpt-export source reads the export zip from ChatGPT's data controls, or the folder you unzipped it to, and indexes each conversation as one memory in a partition called ChatGPT history. It follows the visible reply chain and leaves edited and regenerated side branches out. Available from ModelBrain 0.6.6.

Before you start

Request your export from ChatGPT and download it. You can use the zip as it is or unzip it first.
Use ModelBrain 0.6.6 or later.
Nothing is added until you add it. ModelBrain never guesses a source kind from a file extension, so a folder you already sync as a plain vault is not reindexed by this feature.

Add it

Open the ModelBrain menu, choose Vaults, then Add Source, then ChatGPT export….
A prompt asks whether your export is a zip file. Choose Yes to pick the zip. Choose No to pick the unzipped folder.
ModelBrain indexes it in the background.

From a terminal, connector_sync register <id> <path> --type chatgpt-export does the same. For example: connector_sync register chats ~/Downloads/chatgpt-export.zip --type chatgpt-export. Add --kind-option include_do_not_remember only if you want the conversations described below.

What it reads

The conversations-NNN.json files in the export. An older export with a single conversations.json works too.
One memory per conversation, keyed by its conversation id. The id does not depend on which file holds the conversation or on whether the export is zipped, so a fresh export of the same account leaves unchanged conversations alone.
The text of your messages and the assistant's replies. An image becomes the placeholder [image] and no image data is read.
The title and the creation date, which becomes the memory's date.
If two exports overlap, the most recently updated copy of a conversation wins.

What it skips

Side branches. Only the chain from the last message back to the start is indexed.
Reasoning, tool and system messages.
Conversations marked do not remember. They are skipped and counted. If any copy of a conversation is marked, it stays skipped. Registering with --kind-option include_do_not_remember includes them.
Every other file in the export, such as chat.html, user.json, settings and images. They are never opened.

Limits and safety

A container file over 4 GiB is refused and recorded as a failure, not dropped. Set MODELBRAIN_CONTAINER_MAX_BYTES to change that ceiling. A single member over 32 MiB is refused the same way. Each conversations file is also held in memory while it is read, so one over 256 MiB is refused. A zip is searched to 6 folder levels and up to 4,096 files.

ModelBrain removes a member from memory only after it has read its container to the end. If a file cannot be parsed, every member already indexed from it stays put until the next good pass.

This source is hidden from relay callers. Anything that reaches ModelBrain through the hosted relay, which is how ChatGPT and Grok connect, cannot see what this source imported unless you add its partition (and the partition of any other source of the same kind) to that connection's allowlist yourself. Your chat history is also the most personal thing in this list, so check the partition before you share a ModelBrain store with anyone.

Check that it works

Ask your assistant to run ModelBrain diagnostics. The source reports its kind, members, members_failed and last_container_bytes. For this source it also reports members_skipped_do_not_remember, the number of conversations left out.
Ask a local assistant about something you once discussed in ChatGPT.

If something does not work

The source reports a failure and indexed nothing

The folder or zip held no conversations json file, or one of them was not a JSON list of conversations. Pick the export root again.

A conversation is missing

It may be marked do not remember, or it had no visible text. Check members_skipped_do_not_remember in diagnostics.

Questions

How do I add my ChatGPT export to ModelBrain?

Open the ModelBrain menu, choose Vaults, then Add Source, then ChatGPT export. Say whether you have the zip or the unzipped folder, then pick it. Available from ModelBrain 0.6.6.

Does ModelBrain index every message in a conversation?

It indexes the visible chain of user and assistant text messages. Edited and regenerated side branches, reasoning, tool and system messages are left out, and images appear as the placeholder [image].

What happens to conversations marked do not remember?

They are skipped and counted in diagnostics. You can include them by registering the source with the include_do_not_remember kind option.

Can ChatGPT or Grok read this partition?

No. The ChatGPT history partition is hidden from relay callers, which is how ChatGPT and Grok connect.

Other sources

Each new source kind has its own page.

Claude Code memory → · Mail → · Calendar and contacts → · Bookmarks →

Connect an assistant

Sources feed the same memory these assistants read.

Claude Desktop → · Claude Code → · Cursor →