Search 1,000 AI chats
in 0.1 seconds.
Every decision, every “we tried X but it failed because Y” — searchable forever. Try the demo below, then upload your own ChatGPT data.
This is a demo. Upload your own ChatGPT data to build your palace.
Same MemPalace. Zero setup.
The open-source mempalace engine, hosted for you.
CLI version
pip install mempalace
Python 3.9+ required
ChromaDB runs locally
You manage everything
Web version (this site)
Upload in your browser
No Python, no terminal
Hosted MCP endpoint
Semantic search built in
What you get
96.6% recall (LongMemEval)
Semantic search, not keywords
One MCP URL for any AI tool
Browse your palace visually
3 steps. 60 seconds. Done.
From ChatGPT export to searchable AI memory.
Export your chats
Open ChatGPT Settings → Data Controls → Export
Download zip from email, unzip
Find conversations.json
Upload & embed
Drop the file above. Your browser parses it and generates semantic embeddings (same all-MiniLM-L6-v2 model as the CLI version).
Preview is fully local. Data only uploads when you confirm.
Connect your AI
Get your personal MCP URL. One line to connect Claude, Cursor, or any MCP tool.
✓ Connected. 847 conversations indexed.
What it looks like in practice
$ claude mcp add mempalace https://mempalace.me/api/mcp?token=mp-xxx
✓ Connected. 847 conversations · 12 wings · 67 rooms indexed.
--- next session ---
You: why did we choose JWT over sessions last month?
■ Found in Engineering / Auth & JWT sessions
2026-02-15 · similarity: 0.94
“JWT gives us stateless auth across microservices. Session-based would require Redis for every service. The tradeoff is token revocation — we handle that with short expiry + refresh tokens.”
↑ verbatim from your original conversation
Powered by the MemPalace engine
Semantic Search
Find by meaning, not keywords. all-MiniLM-L6-v2 embeddings + pgvector cosine similarity.
96.6% Recall
Benchmarked on LongMemEval (ICLR 2025). Verbatim storage beats AI extraction.
5 MCP Tools
status, search, list_wings, list_rooms, traverse. One URL for any MCP client.
ChatGPT Parser
Understands ChatGPT export format. Claude and Slack support coming.
Auto-Organize
Conversations sorted into wings (topics) and rooms (sub-topics) automatically.
Your Token, Your Data
Only your personal MCP token can access your palace. Delete anytime.
Questions
How is this different from pip install mempalace?
Same concept, different delivery. The CLI version runs entirely on your machine with ChromaDB. This web version lets you skip Python setup — upload in your browser, get a hosted MCP endpoint. Same embedding model (all-MiniLM-L6-v2), same semantic search quality.
Is my data safe?
Parsing and embedding happen in your browser. When you save, data is stored on our server so your MCP tools can access it. Only your personal token can query your data. You can delete everything from your settings.
What is MCP?
Model Context Protocol — an open standard that lets AI tools access external data sources. One URL connects your palace to Claude, Cursor, ChatGPT (via plugins), and any MCP-compatible tool.
How do I export my ChatGPT data?
ChatGPT → Settings → Data Controls → Export data. You'll get an email with a download link. Unzip it and find conversations.json.
Will you support Claude exports?
Claude and Slack exports are on the roadmap. The open-source mempalace engine already supports them — we're working on bringing that to the web version.
Is this affiliated with the mempalace open-source project?
mempalace.me is an independent hosted version built on the same principles. We link to and credit the original project. The engine is MIT-licensed.
Stop re-explaining yourself to AI.
Upload your conversations. Get a searchable memory palace in 60 seconds.
Free · Based on the MIT-licensed mempalace engine