toolcurrent
Navigation

Explore

MemPalace logo

MemPalace

Open SourceProductivity Last updated: April 9, 2026

MemPalace is a free, open-source local AI memory system giving LLMs persistent, cross-session memory via verbatim storage and semantic search.

ProductivityBest Productivity Software →
Try MemPalace Website

Our General Score

7.4/10
Functionality8.5
Features7.5
Usability5.5
Value9.5
Integrations6.5
Reliability6.5

Plans & Pricing

Use Cases

Coding

8.8

19 MCP tools integrate directly with Claude Code and Cursor, enabling automatic memory saves every 15 messages and pre-compaction hooks that preserve session context before context window compression.

Personal Productivity

8.2

Verbatim storage of Claude, ChatGPT, and Slack exports lets individual users recover decisions, preferences, and reasoning from past sessions without re-explaining context.

Research

7.5

Temporal knowledge graph with validity windows supports structured recall of decisions and milestones over time, though no collaboration or multi-user features exist.

Automation

7.0

MCP tool surface allows agents to query memory programmatically without manual search steps, but setup requires Python environment configuration and MCP server wiring, adding friction for non-developer users.

Platforms

DesktopAPI

Capabilities

Context WindowN/A
API PricingN/A
Image Generation✗ No
Memory Persistence✓ Yes
Computer Use✗ No
API Available✓ Yes
Multimodal✗ No
Open Source✓ Yes
Browser Extension✗ No

Overview

MemPalace is an MIT-licensed AI memory framework created by Milla Jovovich and engineer Ben Sigman, released April 5, 2026. It solves AI session amnesia by storing conversation history verbatim in a hierarchical palace structure — wings, halls, rooms, closets, and drawers — and using ChromaDB vector search and a SQLite temporal knowledge graph to retrieve relevant context on demand. It runs entirely locally with zero API costs, exposes 19 MCP tools for Claude, ChatGPT, Cursor, and Gemini CLI, and scores 96.6% on LongMemEval in raw mode, outperforming paid competitors Mem0 and Zep. It requires Python setup and does not yet support Windows.

Key Features

  • Verbatim storage of all conversation data in ChromaDB without AI summarization or information extraction
  • Hierarchical palace structure organizing memory into wings, halls, rooms, closets, and drawers with 34% retrieval accuracy boost over flat storage
  • 19-tool MCP server for Claude, ChatGPT, Cursor, and Gemini CLI with automatic memory querying during agent sessions
  • Temporal SQLite knowledge graph with validity windows and contradiction detection for entity-relationship tracking across time
  • AAAK compression dialect providing lossless token efficiency for repeated entities at scale, readable by any LLM without a decoder
  • Three mining modes — projects, convos, and general — ingesting codebases, chat exports from Claude/ChatGPT/Slack, and mixed data with auto-classification

Pros & Cons

Pros

  • Zero cost versus paid competitors Mem0 ($19–249/month) and Zep ($25+/month), with equivalent or higher LongMemEval benchmark scores
  • All data remains on-device with no external API calls, providing full data sovereignty for code, business logic, and sensitive conversation logs
  • Startup context loads in 170 tokens, preventing context window bloat that degrades performance with CLAUDE.md-based memory approaches
  • Active community with 23,000+ GitHub stars within 48 hours of launch and open issue tracker with responsive maintainers

Cons

  • Windows not supported as of April 2026; requires Python environment, MCP server configuration, and hook setup, excluding non-developer users entirely
  • Claimed 100% LongMemEval score is contested — the verified raw mode score is 96.6%; the 100% figure required targeted fixes and LLM reranking
  • AAAK compression currently scores lower than raw mode (84.2% vs 96.6%) and the token-saving claims at small scale were acknowledged as incorrect in the README
  • No GUI, no cloud sync, no multi-user collaboration, and no Windows support limit deployment to local Unix/macOS developer environments

Who It's For

Best For

  • Developers using Claude Code, Cursor, or Gemini CLI who need persistent cross-session memory without cloud dependency
  • Privacy-sensitive teams whose conversation logs contain proprietary code or customer data that cannot leave the local machine
  • Indie hackers and bootstrapped founders replacing paid memory APIs to eliminate $200–3,000/year in infrastructure costs
  • AI agent builders who need a callable MCP memory service integrated into multi-agent workflows

Not Ideal For

  • Windows users — no Windows support as of April 2026
  • Non-technical users without Python environment setup capability
  • Teams requiring multi-user shared memory, cloud sync, or a web-based GUI
  • Workflows needing managed SLA, enterprise support, or hosted infrastructure

Audience Scores

Developers9.0

Native MCP integration with Claude Code, Cursor, and Gemini CLI, plus Claude Code auto-save hooks and a 19-tool MCP surface, makes MemPalace purpose-built for developer agent workflows at zero cost versus Mem0 ($19–249/month).

Researchers7.4

Verbatim storage preserves original reasoning rather than AI-extracted summaries, and the temporal knowledge graph supports structured recall across multi-session projects, though no shared team workspace or multi-user access is available.

Product Managers6.8

Can mine Slack exports and chat histories to recover architecture decisions and sprint context, but requires Python CLI setup with no GUI, making it inaccessible without developer support.

Freelancers7.2

Eliminates the cost of paid memory tools ($19–249/month) for solo practitioners running Claude or ChatGPT workflows, but Windows is not yet supported as of April 2026.

Often Used With

Cursor

MCP-compatible AI code editor that connects to MemPalace's 19-tool server for in-session memory queries during development.

Consider These Instead

When Not To Choose MemPalace

Choose Mem0 when you need a managed cloud API with SLA, app-facing REST endpoints, or abstracted summarization-based memory without local setup — pricing starts at $19/month. Choose Zep when you require a conventional hosted memory service with Neo4j-backed knowledge graphs and app-facing APIs without managing local retrieval infrastructure. Choose Letta when the requirement is a full stateful agent runtime rather than a memory retrieval layer.

Integrations

Claude CodeChatgpt (Export Ingestion)CursorGemini CliSlack (Export Ingestion)Ollama

Known Limitations

ecosystem weaknessreliability risklearning curvefeature gap