dsh-memory

DeepSeek Harness starts every conversation from scratch, so user preferences, project context, and critical decisions are lost the moment a session ends. dsh-memory closes that gap: before answering, it automatically recalls relevant memories and injects them into the runtime context; after answering, newly surfaced facts worth keeping are saved to a local JSON store, giving dsh genuine cross-session long-term memory.

The plugin registers five model tools - memory_save, memory_recall, memory_list, memory_update, and memory_delete - backed by a smart search engine that scores results by keyword, tag, importance stars, and recency. Memories rated 3 stars or higher are automatically injected into the context of every request without any manual intervention. A five-level importance scale ranges from ephemeral notes to critical, never-to-be-forgotten decisions. The web UI also gains a dedicated visual settings page where you can browse, search, add, edit, delete, and import or export memories as JSON.

If you rely on dsh day in and day out for project communication, technical decisions, or personal preference management, and want it to remember the things you have already explained, dsh-memory is a lightweight yet practical choice.