Collective Memory

Gives Claude persistent memory across sessions using semantic search. Exposes two tools: remember() to store decisions, learnings, and context with optional project scoping and type tags, and recall() to search that memory bank with natural language queries. Uses OpenAI embeddings and LanceDB under the hood, automatically deduplicates similar content above 95% similarity. Everything lives in a local embedded database at ~/.collective-memory/data, no separate server process. Reach for this when you're tired of re-explaining project context every session or want Claude to actually remember architectural decisions and patterns from previous conversations. Also ships with a CLI for command line memory management.