Remem

Founder’s note

I built Remem because every agent framework I used made memory an afterthought, a summarization hack or a vector store bolted on with no way to see why a memory got retrieved or ask it to forget something. I wanted memory that's a first-class, inspectable part of the system: hybrid scoring that combines relevance, recency, and importance, full visibility into why a memory surfaced, and real control over what gets stored, retrieved, or deleted. I'm building this solo, currently in production hardening — swapping in better embeddings, tightening multi-tenant isolation, fixing the rough edges that only show up once real traffic hits it. The people I most want to reach are developers and small teams shipping production agents who need to trust what their agent remembers, not just demo it. What I'd still improve: deeper framework integrations (LangChain/CrewAI out of the box), a public benchmark page, and a smoother self-serve onboarding flow. Feedback from early users is what will shape all three.