Why Your AI Forgets Everything — and How MemPalace Fixes It

Every AI conversation starts from scratch. Close a ChatGPT or Claude session, and everything
discussed — decisions, reasoning, context — vanishes.

This is the problem MemPalace solves.

## What is MemPalace?

MemPalace is an open-source AI memory system created by Milla Jovovich and Ben Sigman. It gives
any LLM persistent, cross-session memory that runs entirely on your local machine.

Install: pip install mempalace

## The Key Insight

Most AI memory systems (Mem0, Zep, Supermemory) use an LLM to extract "key facts" and throw away
the rest. This is lossy compression — the AI decides what matters, and everything else is gone
forever.

MemPalace takes the opposite approach: store everything verbatim, then make it findable.
Nothing is lost. When you need context from 3 months ago, it's there — the full reasoning,
alternatives considered, and nuance.

## Benchmarks

| System | LongMemEval R@5 | Cost | Local |
|--------|----------------|------|-------|
| MemPalace (hybrid) | 100% | Free | Yes |
| MemPalace (raw) | 96.6% | Free | Yes |
| Supermemory | ~99% | Paid | No |
| Mem0 | ~85% | $19-249/mo | No |
| Zep | ~85% | $25/mo+ | No |

MemPalace is the first system to score a perfect 100% on
LongMemEval. The raw mode — requiring zero API keys — still
beats every competitor.

## How It Works

The architecture borrows from the ancient "memory palace" mnemonic technique:

• Wings — top-level containers for projects or people
• Rooms — specific topics within a wing
• Halls — corridors connecting rooms by memory type (facts, events, discoveries)
• Closets — 30x compressed summaries in AAAK lossless format
• Drawers — original verbatim files (never lost)

The system wakes up with just ~170 tokens and pulls deeper memories on demand.

## Getting Started (5 minutes)

bash
  pip install mempalace
  mempalace init

  Connect to Claude Code:
  {
    "mcpServers": {
      "mempalace": {
        "command": "mempalace",
        "args": ["mcp"]
      }
    }
  }

  Full setup guide: mempalace.tech/guides/setup

  MemPalace vs Mem0

  The two biggest players in AI memory take fundamentally different approaches:

  - MemPalace: verbatim storage, local-first, free, 100% benchmark
  - Mem0: LLM extraction, cloud-first, $19-249/mo, ~85% benchmark

  Mem0 has $24M in funding and enterprise support. MemPalace has higher accuracy and zero cost.

  Detailed comparison: mempalace.tech/compare/mempalace-vs-mem0

  When to Use What

  Choose MemPalace if: you want maximum accuracy, full privacy, zero cost, and local-first
  operation.

  Choose Mem0 if: you need enterprise SLAs, managed cloud infrastructure, and an established vendor.

  ---
  MemPalace is MIT licensed and available on GitHub and PyPI.