deniscartin/llmemoryos

MemoryOS is a self hosted memory system that gives AI models and coding assistants a persistent place to store and recall information, since language models normally forget everything once a conversation ends. Any model or tool, such as OpenAI's models, Claude, Gemini, or coding assistants like Cursor and Windsurf, can connect to it, but none of them talk to the underlying database directly. Instead they go through a Memory API or an MCP server, a standard way for AI tools to call external services, which acts as the single doorway into the system. Rather than being just a place to store searchable text snippets, MemoryOS organizes several different kinds of resources: memories, skills, documents, prompts, tools, and projects, plus a knowledge graph that tracks typed relationships between all of them. It supports many different providers for turning text into the numerical representations search relies on, including a built in option that needs no API key at all, so search works right away without any setup cost. Search itself combines meaning based matching with traditional keyword matching and can rerank results for better relevance. Background workers handle housekeeping tasks like removing duplicate memories, merging related ones, scoring importance, and regenerating summaries. The system also includes an admin dashboard for browsing and managing everything through a web browser, security features like API keys and rate limiting, and typed helper libraries for Python and another popular programming language, making it easier to integrate from your own code. To run it, you need Docker and Docker Compose installed. After cloning the repository and copying an example configuration file, one command builds and starts the entire stack, including the dashboard, the API, and supporting services, all running locally on your own machine or server rather than a third-party cloud service. The project is released under the MIT license.