GitHub radar
HMS: Structured Memory for LLM Reasoning
ShadowWeave’s framework for studying how structuring evidence in a ledger prior to generating answers enhances language models’ ability to reason over long-term and multi-session dialogues.
ShadowWeave’s Holographic Memory System (HMS) is a reproducible long term memory research framework designed for the LongMemEval benchmark, which requires the model to answer questions based on evidence from different previous sessions. It consists of an intermediate evidence ledger (a structured timeline of when, what and where) generated before generating answers, rather than just giving the model a list of retrieved evidence snippets. We also provide a wrapper (wrap_openai()) that can be used with any OpenAI API client, allowing you to automatically keep and look up your memories without modifying your code.
Why a vibe-coder should care
An example of how to implement persistent memory in LLM applications. Simply wrap your model using wrap_openai() and it will start to store and recall useful information from previous interactions. Can be used as inspiration for creating your own AI applications that require long term memory.
▌ More finds