Skip to main content
Bring your existing context into 3ngram from your AI tools. You’ll end up with the same memories, searchable from every MCP client you connect.

​
Supported sources

SourceFormatFlow
ChatGPTconversations.json (data export)LLM extracts memory candidates. You review, then import.
Claudeconversations.json (data export)LLM extracts memory candidates. You review, then import.
Claude CodeMEMORY.md file or zip of ~/.claude/projects/*/memory/Each entry becomes a memory candidate. You review, then import.
All imports are deduplicated against existing memories by content hash, and embeddings are generated automatically so imported memories participate in semantic search.

​
ChatGPT

​
Export your data

  1. Open ChatGPT settings → Data controls.
  2. Click Export data. OpenAI emails a link, usually within a few minutes.
  3. Download the archive and unzip it. Locate conversations.json.

​
Import into 3ngram

  1. Open app.3ngram.ai and go to Settings → Import.
  2. Pick ChatGPT in the source filter and upload conversations.json.
  3. 3ngram processes up to the first 20 conversations and shows extracted candidates.
  4. Review each candidate (type, confidence, source conversation), deselect anything you don’t want, and click Import.
We only process the first 20 conversations by default to keep extraction fast and within per-user LLM budget. If you need more, re-run the import on a smaller slice of the archive or wait for the larger-batch background job (tracked in the roadmap).

​
Claude

​
Export your data

  1. Open Claude settings at claude.ai/settings/data-privacy-controls.
  2. Choose Export data. Anthropic emails a link when the archive is ready.
  3. Download and unzip the archive. Locate conversations.json.

​
Import into 3ngram

  1. Open app.3ngram.ai and go to Settings → Import.
  2. Pick Claude in the source filter and upload conversations.json.
  3. 3ngram linearises each conversation’s flat message array and runs the same extraction flow as ChatGPT.
  4. Review extracted candidates and click Import.
Claude’s export uses a flat chat_messages array per conversation (no branching tree), so nothing is lost in linearisation.

​
Claude Code

Claude Code stores per-project memory at ~/.claude/projects/<project-slug>/memory/MEMORY.md. You can import either a single MEMORY.md or a zip of the entire memory directory.

​
Export your memory

Single project: 
cp ~/.claude/projects/<project-slug>/memory/MEMORY.md ./MEMORY.md
All projects: 
cd ~/.claude/projects && zip -r ~/claude-memory.zip */memory/

​
Import into 3ngram

  1. Open app.3ngram.ai and go to Settings → Import.
  2. Pick Claude Code in the source filter and upload your .md file or zip.
  3. 3ngram parses each top-level section heading as a memory entry.
  4. Review candidates and click Import.
After the initial import, install the 3ngram Claude Code plugin so future memories are captured directly into 3ngram instead of MEMORY.md.

​
Troubleshooting

​
File size limits

  • ChatGPT / Claude extraction: 10 MB per upload.
  • Claude Code and document imports: 50 MB per upload.
If your export is larger, split conversations.json into chunks (one top-level JSON array per chunk) and upload them separately.

​
Conversation cap

Extraction defaults to 20 conversations per upload. You’ll see conversations_processed in the API response — anything above the cap is silently dropped from this run. Re-upload a filtered archive to process the rest.

​
Extraction rate

Extraction runs sequentially through conversations, ~1-2 seconds per chunk. A 20-conversation export typically finishes in under a minute. If it stalls, check Settings → Billing for your monthly LLM budget — extraction fails gracefully (but noisily) when you hit your cap.

​
No memories found

If extraction returns zero candidates:
  • The conversations may be pure Q&A with no decisions, commitments, preferences, or patterns worth remembering.
  • System messages and code snippets are intentionally skipped.
  • Try uploading a different export window that contains planning or decision-making sessions.

​
Export

Download your entire workspace as a ZIP archive at any time:
  1. Go to Settings → Export.
  2. Click Download ZIP.
The export includes all documents, memories, and metadata. Available regardless of subscription status. Your data is always yours.