quantumregen/roguecore

RogueCore is an experimental on-device memory system that stores knowledge using a different approach than most modern AI tools. Instead of a traditional database or neural network weights, it encodes concepts as wave interference patterns in a complex-valued mathematical field, an approach the author describes as inspired by how holograms store information in distributed patterns rather than as discrete records. The library is built on PyTorch. The central idea is a stable core_self anchor, a protected identity at the center of everything the system knows. All other knowledge is organized around this anchor, which the author says gives the system a persistent sense of identity across sessions. You interact through simple commands. Teach adds a concept by name and description, ingest_text reads a file and strengthens existing concepts rather than adding noise, ask retrieves information by associative recall, and ponder and reflect trigger internal exploration modes. Everything saves to disk and reloads across sessions. The system is designed to run on a personal laptop or edge device. The README describes it as extremely low resource usage once loaded, after a 1024 by 1024 complex wave field is initialized. It has no external dependencies beyond PyTorch and is described as fully inspectable, meaning you can see exactly what the core knows and how strongly each concept is stored. The author frames it as a deliberate alternative to large, opaque AI systems: small, transparent, personally owned memory. Current limitations noted in the README include being primarily pattern-association rather than full reasoning, and mostly lexical rather than semantic concept binding. Written in Python.