Hollywood actress Milla Jovovich has teamed up with @ bensig to develop the open-source AI memory tool MemPalace! The inspiration for this tool comes from the ancient Greek technique of "memory palace" (storing information in imagined spatial locations), but it is applied to AI agents. It enables AI to remember conversations, decisions, knowledge, and other content for a long time, running entirely locally without the need to call external APIs (for offline use). Main features and highlights: -On the industry standard long memory evaluation LongMemEval, it achieved R @ 5 100% (mixed mode) and a raw score of 96.6%, which is currently the highest publicly available result (whether paid or free tool). -By using ChromaDB, SQLite knowledge graph, and custom AAAK lossless compression technology, the compression efficiency can reach about 30 times, making memory more efficient. -Equipped with advanced features such as conflict detection, specialist agents, and MCP tool integration. -Completely open source, already on GitHub, welcome everyone to try, comment, fork or contribute code, and join their Discord community. -Milla Jovovich personally participated in the development and even posted a related video introduction on Instagram. This' Memory Palace 'tool achieved 100% (mixed) and 96.6% original scores on the LongMemEval benchmark test, surpassing all publicly available results to date! Completely run locally, no API required, supports conflict detection and expert agents, and achieves 30 times efficiency with custom compression. What does this mean? Traditional AI (such as ChatGPT, Claude) is prone to "forgetting" previous details, decision-making processes, or cross conversation context during long-term conversations. MemPalace solves this problem by simulating the ancient Greek "memory palace" technique, storing all conversations, knowledge, and decisions completely on a local computer and structuring them into "palaces, halls, and rooms", allowing AI to accurately recall past content. Core advantages: *Fully local operation: No need for any API keys or cloud services, completely offline, privacy and security. *Efficient compression: Using custom AAAK lossless compression technology, memory efficiency is improved by about 30 times, and thousands of conversations can be stored without taking up too much space. *Top performance: On the industry standard long memory benchmark LongMemEval R @ 5, the original mode achieved 96.6% (without any external LLM), and the hybrid mode (optional rerank) achieved 100% (first perfect score), which is currently the highest publicly available record. *Advanced features: automatic linking of related topics, conflict detection, expert agents, support for MCP protocol (can be called with Claude and other model integration tools). *Open source and free: MIT license, with tens of thousands of stars quickly accumulated on GitHub. Welcome to fork, contribute, and join Discord in the community. This is a rare Hollywood celebrity who has deeply participated in AI open source projects, and Milla herself even took selfies to explain the inspiration for the video (after accumulating thousands of AI conversations, she found that existing tools always "help you decide what to remember" and discard important context). How do users use MemPalace? 1. Easy installation (Python environment): *Execute instruction: pip install memspace (only requires Python 3.10+, relies on ChromaDB and PyYAML, no other complex packages). *Initial setup: mempalace init~/Your project directory (will guide you to create a memory palace structure). 2. Mining and importing data: *Use CLI commands to automatically scan your project folder, chat history, code, etc., and convert them into structured memory (supports AAAK compression). 3. Integrate into AI workflow: *Connect Claude, GPT, Llama and other models through MCP (Model Context Protocol) in a JSON-RPC manner. *Call tools in the conversation, such as mempalace_stus to view an overview of the palace, read to extract specific memories, etc. (about 20+tools in total). *Completely local, suitable for use with offline LLMs such as Ollama or LM Studio. For detailed installation and commands, please refer to the official GitHub: https://(GitHub)/villa jvovich/mempalace Actual use cases: *Long term personal/professional knowledge base: Writers, researchers, or creators can save months of conversations, inspirations, and research notes, and AI can accurately reference them next time without repeatedly explaining the background. *Software development assistant: Programmers remember all project history, debug decisions, and architecture discussions, and AI can recall how we handled this bug last time over several weeks. *Privacy sensitive work: Lawyers, physicians, or internal teams within companies require offline, secure memory systems to avoid uploading data to the cloud. *Multi agent system: allowing different "expert agents" to share the same memory palace, automatically linking cross topic content, and improving the efficiency of complex tasks. *Daily productivity: Students and freelancers store study notes, meeting minutes, and ideas as a memory palace, while AI becomes an unforgettable personal secretary.