96.6% recall · $0 · Runs on your machine

Your memory.
Your machine.
Your rules.

You've spent 1,000 hours talking to AI. Every decision, every “we tried X but it failed because Y” — gone. MemPalace keeps everything and makes it searchable.

Drop your conversations.json here

ChatGPT export · parsed locally · nothing leaves your browser

How to get your ChatGPT data

1. ChatGPT Settings → Data Controls → Export

2. Download zip from email, extract it

3. Find conversations.json and drop it above

Open source · MIT license · No API keys needed

Your AI conversations are a knowledge goldmine.
You're letting it evaporate.

1,000+
hours of AI conversations in 6 months of daily use
19.5M
tokens of decisions, debates, and discoveries
0%
retained across sessions by any AI tool today
Every time you re-explain your tech stack, re-debate a decision you already made, or search through chat history for 20 minutes — that's the cost of evaporation.

Other tools let AI decide what to remember.

MemPalace keeps every word. That's why it's 2x more accurate.

MemPalaceMem0ZepChatGPT Memory
Recall accuracy96.6%49-85%63.8%N/A
Monthly cost$0$19-249$25+w/ subscription
Data locationYour machineCloudCloudOpenAI
Information lossZeroHigh (AI extracts)MediumExtreme
Works withAny LLMAPI onlyAPI onlyGPT only
Benchmarks from LongMemEval and ConvoMem. MemPalace uses local ChromaDB embeddings — no API calls.

What it looks like in practice

claude code

$ claude mcp add mempalace https://mcp.mempalace.me/u/abc123

✓ Connected. 847 conversations · 12 wings · 67 rooms indexed.

--- next session ---

You: why did we choose JWT over sessions last month?

■ Found in wing_projectA / room_auth-architecture
2026-02-15 · Claude conversation · 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. Not a summary. Not an AI extraction.

What's inside

Conversation Parser

ChatGPT JSON, Claude exports, Slack. Auto-detect projects, people, topics.

96.6% Semantic Search

Find by meaning, not keywords. ChromaDB embeddings, local, no API.

Knowledge Graph

Entities, relationships, timelines. Auto-extracted, queryable, temporal.

19 MCP Tools

Search, entity lookup, diary, navigation. Any MCP client, one URL.

AAAK Compression

30x token compression, zero info loss. 170 tokens to load months of context.

Fully Local

Your data never leaves your machine. MIT license. pip install mempalace.

Questions

How do I get my ChatGPT data?

Settings → Data Controls → Export. You get a zip via email. Upload conversations.json.

Is my data safe?

MemPalace runs 100% locally. No cloud, no API calls for search. Your data stays on your machine. Open source — audit every line.

How is it free and better than paid tools?

Other tools spend compute on AI extraction (which loses info). MemPalace skips extraction and uses raw text + structure. Less compute, better results, $0.

What is MCP?

Model Context Protocol — lets any AI tool access your memory. One URL connects Claude, ChatGPT, Cursor. 19 tools out of the box.

Can I self-host the web version?

Engine: pip install mempalace (MIT). Web version Docker image coming. Or use mempalace.me for hosted access.

Stop re-explaining yourself to AI.

Upload your conversations. Give your AI a memory.

Free · Open source · MIT license · Your data stays yours