MemPalace Logo MemPalace MCP Hub GitHub Repo

Your AI Remembers. Locally.

Use the official MemPalace package (v3.3.2, released April 21, 2026) with MCP support for Claude, ChatGPT, Cursor, and Gemini. Keep memory local-first with 96.6% raw R@5 and 98.4% held-out hybrid R@5 on LongMemEval.

100% Local $0 Subscription 29 MCP Tools (official) 96.6% LongMemEval Score
View Setup Guide

Why a Hollywood Actress Started Coding AI Tools

Milla Jovovich (Alice in Resident Evil) noticed a glaring issue with LLMs: every new session suffered from amnesia. Existing memory tools used AI to decide what to forget, discarding the very reasoning she needed most.

Partnering with developer Ben Sigman, they built MemPalace. Their radical approach? Store everything verbatim and make it instantly findable using a "memory palace" metaphor.

"I wanted my AI to remember the way I remember — not just the conclusions, but the journey. The alternatives I considered... That's what existing memory systems throw away."
Milla Jovovich & Ben Sigman
25.4K+
GitHub Stars
100%
LongMemEval
$0
API Cost
Local
Zero Cloud

Three Ways to Run MemPalace

The DIY Way

$0/mo

Self-host using our MCP guides. Run the Python server and ChromaDB locally.

Read Docs

The Cloud Way

€10/mo

Don't want to run a local server? Use the managed backend hosted by the community.

In Private Beta
Enterprise

Business Automation

Custom

Want to give your WhatsApp chatbots and CRM long-term memory? We integrate MemPalace with n8n and Pabbly for service businesses.

Book VijayaTech Labs

Latest Official Updates (April 2026)

Release: MemPalace v3.3.2 was published on April 21, 2026.

Package status: PyPI currently lists 3.3.2 as latest.

MCP tools: Official package docs now describe 29 MCP tools.

Safety notice: Official docs warn that mempalace.tech is not an official domain.

Sources: GitHub Releases · PyPI · Official Docs

Connect MemPalace to Your Workflow

Claude Code

Plugin + MCP workflow in official docs.

ChatGPT

Via MCP bridge, full tool access.

Cursor

Code editor with persistent memory.

Gemini CLI

Auto server + save hooks.

n8n / Make

Search memories in automated workflows.

Local LLMs

Wake-up context injection (Llama, Mistral).

How MemPalace Compares

Official benchmark reporting now emphasizes reproducible numbers: 96.6% raw R@5, 98.4% held-out hybrid R@5, and reranked runs reported as ≥99% rather than "100%" headlines.

Reproducible local-first benchmark
Tool API / Memory System Recall (R@5) Runs Locally External API Required Monthly Cost
MemPalace (Raw) 96.6% Yes No $0
MemPalace (Hybrid, held-out) 98.4% Optional Optional $0
Mem0 ~85% No Yes $19–249
Zep ~85% Enterprise Only Yes $25+
Letta N/A No Yes $20-200

How the Palace Works

Wing

Project / Person

Room

Topic / Context

Closet

AI Summary

Drawer

Raw Verbatim File

Core Tool Snapshot (19 of 29)

  • mempalace_status
  • mempalace_list_wings
  • mempalace_list_rooms
  • mempalace_get_taxonomy
  • mempalace_search
  • mempalace_check_duplicate
  • mempalace_get_aaak_spec
  • mempalace_add_drawer
  • mempalace_delete_drawer
  • mempalace_kg_query
  • mempalace_kg_add
  • mempalace_kg_invalidate
  • mempalace_kg_timeline
  • mempalace_kg_stats
  • mempalace_traverse
  • mempalace_find_tunnels
  • mempalace_graph_stats
  • mempalace_diary_write
  • mempalace_diary_read

Live in 60 Seconds

pip install --upgrade mempalace
mempalace init ~/projects/myapp
mempalace mine ~/projects/myapp
claude mcp add mempalace -- python -m mempalace.mcp_server

What the Community is Saying

3.1K+
Forks
158+
Open Issues
W
Wayne Sutton
@WayneSutton

"Milla Jovovich launching an AI memory system was NOT on my 2026 bingo card."

B
Brian Roemmele
@BrianRoemmele

"We tested MemPalace at The Zero-Human Company and deployed it to 79 employees. This is a masterpiece."

Y
HackerNews
denysvitali

"Yes, that Milla Jovovich. Missed opportunity to call it Resident Eval."

Trending Feature : Import support for Cursor, Copilot, Codex, Windsurf
View Protocol

Calculate Your Token Savings

See how much you save by moving amnesic context to local persistent memory.

10M
1 Million 50 Million
Standard LLM Context
$150.00
API In-Context
Mem0 API (Cloud)
$50.00
SaaS Subscription
MemPalace.in
$0.00
100% Local Storage

Frequently Asked Questions

What is MemPalace?
An open-source, local AI memory system that uses a memory palace metaphor (wings, rooms, closets, drawers) to store and retrieve agent conversations.
Does MemPalace require an API key?
No. MemPalace runs 100% locally using ChromaDB and SQLite, requiring zero cloud dependencies.
What is the LongMemEval score for MemPalace?
In raw verbatim mode, MemPalace scores 96.6% R@5 on 500 questions, independent of LLM APIs.
How do I connect MemPalace to Claude?
For MCP clients, run claude mcp add mempalace -- python -m mempalace.mcp_server. For plugin setup, follow the current Claude Code Plugin page in the official docs.