Show HN: SelMem – selective reconstructive memory for LLMs
Category: ai-ml
Tags: llm, memory, rust, research, reconstruction
Score: 6.0/10 (Innovation: 7, Technical: 7, Documentation: 6, Utility: 4)
SelMem is a Rust library implementing selective reconstructive memory for LLM entities, featuring encode/judge/night/snapshot cycles that shape future context through path-dependent memory rather than simple logging. It's an ambitious research project exploring whether memory selection and reconstruction can induce behavioral divergence and creativity in LLMs, with a hypothesis about two models developing different pasts over time.
Target audience: AI researchers and LLM application developers interested in agent memory architectures
Repository: https://github.com/jbsalles/Selmem · Rust · MIT · 10 stars
View on Hacker News