mindwright gives Claude Code persistent, per-agent memory. It captures what happens during a session, distills it into long-term facts via a background consolidation pass, and surfaces only the relevant ones on demand through fully local retrieval (SQLite + on-device embeddings — no API calls). On a fresh install it bootstraps from your existing project transcripts, so recall is useful from day one. Multi-agent aware: each agent keeps its own memory with role-scoped recall.