About HumemAI
HumemAI started from Taewoon Kim’s PhD thesis, “A Machine With Human‑Like Memory”. The core idea: glue together three pillars—machine learning, knowledge graphs, and SQL/NoSQL databases—and add a human aspect by modeling an agent’s memory as graph‑structured state that persists across SQL/NoSQL, learns, and adapts.
From toy examples to real applications, the thesis explored memory as a knowledge graph, reinforcement learning for memory management policies, and persistence in databases so knowledge isn’t lost. That trajectory became HumemAI.
Today, users interact through an LLM while HumemAI performs hybrid retrieval—vector similarity plus relationship‑aware queries over graph‑structured memory—to ground answers in a continuously evolving knowledge graph layer. It powers agentic workflows: natural to use, explainable by design, and reusable across tasks.
We’re open‑source first. The stack—SQL and NoSQL databases (relational, graph, vector, key‑value, wide‑column, document, search), analytics, and LLMs—is built with permissive licenses. Bring your data in any format—tables, graphs, documents—and HumemAI stores it wisely across the right engines. Anyone can self‑host HumemAI for free. We also offer managed hosting (SaaS/PaaS) and tailored ontology design to fit your data.
Our research focus keeps us ahead: integrating ML with graphs and databases to scale reasoning, improve retrieval, and automate schema/ontology learning. That’s how we move beyond short‑lived hype to durable capabilities.