Import Your Entire ChatGPT History in 60 Seconds

You have hundreds — maybe thousands — of ChatGPT conversations. Decisions you made, ideas you explored, code you debugged, strategies you refined. All of it trapped inside OpenAI's interface with no way to search across conversations. Here is how to get it all into REM Labs, fully indexed and searchable, in under a minute.

The export contains your full message history, including timestamps, conversation titles, and both your messages and the assistant's responses. No conversations are excluded. If you have been using ChatGPT since 2022, this file can be surprisingly large — 50MB or more for heavy users — but REM handles files of any size.

Option A: Drag and Drop

Open the REM Console and drag your conversations.json file directly onto the page. The importer detects the ChatGPT format automatically, parses every conversation, and stores each message turn as a separate memory. You will see a progress indicator as memories are indexed. For a typical history (200-500 conversations), this takes 10-30 seconds.

Option B: API

If you prefer to script the import or are integrating into a workflow, use the import endpoint directly:

The response confirms how many memories were extracted and stored:

The imported count reflects the number of distinct memory units created — not the number of conversations, but the number of individual conversation turns that contained meaningful content. Empty messages and system prompts are filtered out automatically.

Try queries like:

• "What did I ask about React hooks?" — finds every conversation where you discussed React hooks, ranked by relevance
• "Summarize my conversations about the product launch" — synthesizes information from multiple conversations into a single answer
• "What was that Python script I wrote for parsing CSVs?" — retrieves the exact conversation turn where you wrote or discussed it
• "What decisions did I make about the database schema?" — pulls together decisions spread across different conversations over time

You can also search programmatically. The recall endpoint returns relevant memories ranked by score:

Or use the full-text and semantic search endpoint for more control:

What Happens After Import

The import is not just storage — it is the starting point for everything REM does with your data.

Entity extraction runs automatically. People, projects, technologies, preferences, and decisions mentioned in your ChatGPT history are extracted and added to your knowledge graph. This means you can query by entity ("What have I discussed with Sarah?") not just by keyword.

The Dream Engine includes your imported history in its nightly consolidation. Patterns across your ChatGPT conversations — recurring topics, evolving decisions, implicit themes — surface in your Morning Brief.

Cross-source connections emerge when your ChatGPT history intersects with other connected sources. If you discussed a project in ChatGPT and that same project appears in your Gmail or Notion, REM links them automatically. Your fragmented context becomes a single, searchable knowledge graph.

FAQ

How long does the export take from OpenAI?

Usually 5-15 minutes. OpenAI sends a download link to your email. For very large accounts, it can take up to an hour.

Is my data private?

Your memories are encrypted and isolated in your own namespace. We do not train on your data, share it, or use it for anything other than serving your queries. See our privacy policy for specifics.

Can I import Claude or other AI histories too?

Yes. The import endpoint accepts data from ChatGPT, Claude, and generic JSON/text formats. For Claude, export your conversation history and use "source": "claude" in the import call. For any other format, use "source": "text" and pass an array of content strings.

Does it count against my store limits?

Each imported memory turn counts as one store operation. The free tier includes 100 stores per month. Pro ($29/mo) includes 10,000 stores — more than enough for most ChatGPT histories. If your export is very large, the import batches automatically and counts against a single billing cycle.