← All posts · 2026-04-26
How to import your ChatGPT history into Moss
Most ChatGPT users have months or years of conversation history sitting in their account. That history is the closest thing to a thinking diary most people have. When you start with Moss, you can either build memory from scratch (chat naturally, Moss learns) — or you can import the ChatGPT history you already have and skip ahead.
This walkthrough covers the import. Total time: about ten minutes, mostly waiting for OpenAI to email you the export.
What you'll get
Once your ChatGPT export is uploaded to Moss:
- Every conversation you've had with ChatGPT is parsed into semantic units (decisions, findings, preferences, facts).
- Those units are indexed into your personal Moss memory with embeddings + entity links.
- Any future Moss conversation can retrieve from this history. Asking "what was that recipe I liked?" pulls from your ChatGPT chat from three months ago, not just from chats you've had with Moss directly.
- Moss preserves the original conversation thread so you can drill into the source if needed.
It is not lossy. Your ChatGPT chats are the ground truth; Moss layers structured memory on top.
Step 1 — Request your ChatGPT export
Open ChatGPT. Click your profile picture (top right). Go to Settings → Data Controls → Export data. Click Export.
OpenAI sends you an email with a download link within a few minutes (sometimes longer for very large accounts). The link expires after 24 hours, so don't queue this for later.
The download is a ZIP file. Unzip it locally. The file you want is conversations.json — that's the canonical export of every chat you've had.
You can ignore the rest of the ZIP for the import (it contains DALL·E images, system files, and a static HTML viewer; none of it is needed for Moss).
Step 2 — Sign up for Moss (or sign in)
Moss is free to start. Sign up at mossmemory.com using Google, Apple, or an email magic link. No credit card.
Document upload is a Sprout-tier feature ($25/mo). The upload route is gated for cost reasons — parsing tens of thousands of messages costs us real OpenAI / Anthropic tokens. If you're on Free, the import is the moment to upgrade. You can downgrade later.
Step 3 — Upload conversations.json
Once you're on Sprout or above:
- Go to your Moss chat (mossmemory.com/chat).
- Click the upload button (paper-clip icon on the chat input).
- Select conversations.json from your unzipped ChatGPT export.
- Moss recognises the format and starts processing.
You'll see a progress indicator. Processing takes anywhere from a few minutes (small history, hundreds of messages) to maybe an hour (years of heavy ChatGPT use, tens of thousands of messages).
Moss does the work in the background. You can close the tab — it doesn't have to stay open. You'll get a notification when the import is complete.
What's actually happening behind the scenes
For each conversation in your export:
- Moss splits it into individual exchanges (user message + assistant response).
- The scribe agent reads each exchange and extracts semantic units — decisions you made, things you preferred, facts about your life and work, contradictions to resolve.
- Each unit gets an embedding (256-dimensional vector) and entity links (people, places, projects, topics).
- Units are de-duped against existing memory so re-imports don't double-count.
- The contradiction-aware curator runs after the import to resolve any conflicts ("I was vegetarian in 2024 → I'm not anymore in 2026").
The output is a personal knowledge graph with full retrieval, ready to use on your next message.
Step 4 — Verify the import
Try a few questions you'd expect Moss to know about, based on your ChatGPT history:
- "What's the project I've been working on most this year?"
- "What did I say about [person/topic] back in [month]?"
- "Summarise what I've learned about [topic] from my conversations."
Moss should retrieve from your ChatGPT history and ground the response in actual things you said. If Moss says "I don't have any record of that" and you know you discussed it in ChatGPT, the import probably hit a parse error on that conversation — email hello@mossmemory.com and we'll dig in.
Updating with new ChatGPT history
ChatGPT keeps generating chats while you use both. To keep Moss in sync, re-export from ChatGPT every few weeks/months and re-upload. Moss de-dupes against existing memory, so you only pay the parse cost on new conversations, not the whole history.
We're working on an automatic sync option (OAuth into ChatGPT, periodic pull). Until then it's manual. Sorry — OpenAI's export-only API limits us here.
Common gotchas
- Very large exports can take a while. If you've used ChatGPT heavily for two years, expect ~30-60 minutes of processing. Run it overnight if it's a lot.
- Multiple ChatGPT accounts — Moss imports per Moss account, not per ChatGPT account. If you have multiple OpenAI accounts you want to merge, upload each export separately to the same Moss account.
- Voice chats — ChatGPT voice transcripts are included in the export and Moss imports them as regular messages. Quality depends on OpenAI's transcription accuracy.
Next: import Claude too
The Claude export format is similar — JSON dump from your Anthropic account. Moss handles it the same way. Step-by-step Claude import guide is coming — for now, upload claude_export.json the same way and Moss will recognise it.
The combined import (ChatGPT + Claude + Gemini exports) gives you a single memory layer across every AI you've ever talked to. That's where Moss starts to feel like the personal AI you've actually wanted.
Sign up for Moss · See the benchmark · Pricing
Try Moss · Blog · Home · Benchmark