Question for people building/testing OpenHands agent workflows: Would a local memory layer be useful if it only stored project context after explicit human approval? I’m testing Memoria, a Windows-first local beta for approval-gated project memory across AI coding tools/MCP clients. The goal is to reduce re-explaining project context without silent memory writes or cloud storage. Fake demo data only, no private code/API keys/source access. If this is relevant to your workflow, open to giving blunt feedback?