MEMM

MEMM is an open-source, AI-native application designed to serve as a persistent, structured memory system for AI tools. It addresses the challenge of AI "amnesia," where large language models and AI assistants repeatedly lose context between sessions, requiring users to re-explain information and maintain redundant knowledge across different tools. MEMM captures and organizes reasoning, conventions, and project knowledge in plain text Markdown files with YAML frontmatter, making the memory transparent, editable, and versionable by the user.

The platform employs a scoring engine based on six signals—BM25, semantic similarity, graph relationships, recency, importance, and frequency—to rank and tier memories for each query. This approach aims to deliver high retrieval precision and context accuracy, while reducing token usage and latency. MEMM's engine operates with sub-millisecond query latency, and its tiered memory system ensures that only the most relevant information is injected into AI queries, avoiding overstuffed or irrelevant context. A governance layer tracks the health of the memory, surfacing stale entries, contradictions, and redundancies, and providing suggestions for consolidation and improvement over time.

MEMM is designed for engineers and users who work with AI tools such as ChatGPT, Claude, Cursor, Codex, and local LLMs, allowing them to connect their AI assistants to a single source of structured knowledge via an MCP server. This eliminates the need to manually synchronize knowledge across multiple platforms and provides a unified, evolving memory accessible to all connected AI tools. The system supports categorizing knowledge as entities, concepts, sources, or syntheses, enabling AIs to reason over structured ontologies rather than flat text.

The application is available for Mac, Windows, and Linux, and is built to be local and portable, ensuring that all knowledge remains owned and controlled by the user. MEMM does not rely on databases, embeddings, or black-box retrieval, instead prioritizing transparency and user ownership. Its open-source nature and focus on context engineering position it as a tool built specifically for the needs of the AI era.