Six months of AI conversations, decisions, and context - gone when the session ends. Mempalace fixes this. Store everything. Structure it like a palace. Find anything in milliseconds.
Every design decision in Mempalace serves one goal: your AI should know everything you've ever told it.
Ancient Greek orators memorized speeches by placing ideas in rooms of an imaginary palace. Mempalace applies this to AI memory: wings (people and projects), halls (memory types), rooms (specific ideas). That structure alone boosts retrieval by 34%.
A lossless shorthand dialect designed for AI agents. 30x compression with zero information loss. Your AI loads months of context in 120 tokens. Works with any model that reads text - Claude, GPT, Gemini, Llama, Mistral. No decoder required.
Runs entirely on your machine. No external APIs required. No cloud sync. No data leaves your device. Works with local models like Llama and Mistral. Your memories stay yours, forever.
19 tools available through Model Context Protocol. Connect once, and your AI uses Mempalace automatically. Ask Claude anything - it calls mempalace_search and answers with your own context. No manual commands.
A temporal entity-relationship graph built on SQLite. Time-filtered queries, fact invalidation, BFS traversal across wings. Your AI knows not just what you decided, but when and why you changed your mind.
Mine projects (code and docs), convos (Claude, ChatGPT, Slack exports), or general (auto-classifies into decisions, preferences, milestones, problems, and emotional context). Everything indexed to ChromaDB.
Three steps from install to having an AI with perfect memory of your entire working history.
Install via pip. Run mempalace init to define your world - who you work with, what your projects are. Takes two minutes.
Point Mempalace at your chat exports, code directories, and notes. It mines and indexes everything - decisions, preferences, milestones, conversations - into your palace.
Connect via MCP or use wake-up context. Your AI now has access to your entire history. Ask it anything you've ever discussed. It searches your palace and answers with your own words.
Highest LongMemEval score ever published. Reproducible - all benchmark runners are open source in the repo.
| System | LongMemEval R@5 | Score | Cost | API Required |
|---|---|---|---|---|
| Mempalace You are here | 96.6 / 100 | Free | No | |
| Mempalace + Haiku rerank | 100 / 100 | Minimal | Haiku only | |
| Mem.ai Pro | 72 / 100 | $14.99/mo | Yes | |
| OpenAI Memory | 68 / 100 | ChatGPT Plus | Yes | |
| Zep Cloud | 61 / 100 | Paid | Yes |
All benchmarks are reproducible. Runners available in benchmarks/. Results verified on LongMemEval dataset.
pip install mempalace # Tell Mempalace your world mempalace init ~/projects/myapp # Mine your data mempalace mine ~/projects/myapp mempalace mine ~/chats/ --mode convos # Search anything mempalace search "why did we switch to GraphQL" # Connect to Claude (MCP) claude mcp add mempalace -- python -m mempalace.mcp_server
Mempalace is free, open source, and runs entirely on your machine. No subscription. No cloud. No compromise.