ReMe

This is actually the kind of tooling AI developers have been cobbling together badly for months, so having it in one framework is refreshing. ReMe combines file-based memory for structured long-term persistence with vector-based retrieval for semantic matching, which means you're not choosing between perfect recall and context-aware search anymore. The context window problem is real and it's been making AI agents useless for anything requiring actual memory, so this dual approach is low-key the move. It's open source on GitHub, which tells me the agentscope-ai team built this because they actually needed it — and that's always a better signal than VC-backed hype.